2027 PROJECT

View 2027 PROJECT flipbook.

C E N T R E O F E X C E L L E N C E Artificial Intelligence Industry Problem Statements & Use Cases — U S E C A S E S P R O V I D E D B Y Knowledge Partner – Cognizant Real-world industry challenges, framed for AI-driven solutions by students and faculty. A C A D E M I C Y E A R 2026 – 2027 R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) RSM NAGAR, KAVARAIPETTAI – 601 206 Affiliated to Anna University, Chennai / Approved by AICTE, New Delhi / Accredited by NAAC with A+ Grade / ISO 21001:2018 Certified Institution

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science Cognizant Hack to Hire 2027- Academic Year 2026-2027 Contents Team Project Faculty mentors Page Summary Gartner Hype Cycle 2026: project-wise summary 4 Team 1 – Hackstreak Risk Adjustment and HCC Suspecting Assistant Dr. A. Ponmalar, Ms. E. Thenmozhi 9 Team 2 – Prime Syndicate Data Sentinel – Claims and Authorization Data-Quality Anomaly Monitor Dr. N. R. Gladiss Merlin, Ms. R. Devi 17 Team 3 – Cognivolve ClaimCare – AI-Powered Claims and Authorization Data Quality Intelligence Dr. A. Ponmalar 26 Team 4 – Enfourcers SDOH-Driven Member Risk and Intervention Prioritization Mr. M. Stanlywit 35 Team 5 – Innovatex CareSync – SDOH-Driven Member Risk and Intervention Prioritization Dr. M. Shanthi, Mr. Satheesh Abraham Leo 43 Team 6 – Tech Connect Provider Intelligence Engine – Network Adequacy and Access Intelligence Mr. S. Nagendiran 52 Team 7 – Tech Vanguard Meridian – AI-Powered Healthcare Claims Risk Detection and Intelligent Investigation Dr. P. Shobha Rani, Dr. B. Muthazhagan 61 Team 8 – Trailblazers Value-Based Care Contract Performance Command Center Ms. C. Vimala Josphine, Mr. V. K. Sabarirajan 69 Team 9 – Techno Squad RightPath – Avoidable Emergency Department Utilization Navigator Dr. G. Manikandan 77 Team 10 – CodeCrafters Claims Fraud, Waste and Abuse Risk Detector Dr. P. Shobha Rani 85 Team 11 – Fantastic Four Prior Authorization Triage and Policy Companion Dr. R. Priyanka Pramila 93 Team 12 – Tech Titans Prior Authorization Triage and Policy Companion Dr. S. Srijayanthi, Ms. B. Mythili 101 Team 13 – InsightHive AI-Driven Healthcare Data Quality and Anomaly Monitoring System Ms. C. Vimala Josphine, Dr. A. Vijayaraj 109 Team 14 – Nexora Nexora – AI-Assisted Prior Authorization Triage and Policy Companion Dr. R. Priyanka Pramila 118

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science SUMMARY GARTNER HYPE CYCLE 2026: PROJECT-WISE POSITION The fourteen projects in this booklet were carried out by students of the Department of Artificial Intelligence and Data Science in August 2026, on healthcare payer use cases given by Cognizant for the 2027 batch. Each project has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026 to show where its main technology stood in industry at that time. The closest Gartner technology profile for each project was identified from its final presentation, and four attributes were recorded: Hype Cycle phase, maturity, market penetration and time to mainstream adoption. Gartner's benefit rating has been left out, as the use cases came from Cognizant and their business value was not assessed. Position of the Projects on the Hype Cycle Figure: Hack to Hire 2027 projects on the Gartner Hype Cycle for AI, 2026 Seven teams (1, 4, 5, 6, 8, 9 and 10) built decision-support systems that fall under decision intelligence, still approaching the Peak in 2026. Three prior-authorization and fraud teams (7, 12 and 14) built multi-agent systems, placing them under AI agents at the Peak of Inflated Expectations. The three data-quality monitors (Teams 2, 3 and 13) relate to AI-ready data, which had just passed the Peak. Team 11 used retrieval-augmented generation, under generative AI in the Trough of Disillusionment, where the focus had moved to reliability and grounding.

How to Read the Hype Cycle The vertical axis shows how much attention and expectation a technology attracts, and the horizontal axis shows its progress over time. Gartner divides the curve into five phases. • Innovation Trigger. A new technology attracts attention, but few usable products exist yet. • Peak of Inflated Expectations. Publicity runs ahead of proven results. • Trough of Disillusionment. Interest falls when early implementations do not deliver what was promised. • Slope of Enlightenment. Benefits become clearer and improved products appear. • Plateau of Productivity. The technology is in mainstream use and its value is well established. Project-wise Summary Team Project Gartner 2026 technolog y Hype Cycle phase Maturit y Penetr ation Time to mainst ream Team 1: Hackstrea k Risk Adjustment and HCC Suspecting Assistant Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years* Team 2: Prime Syndicate Data Sentinel – Claims and Authorization Data- Quality Anomaly Monitor AI-ready data Past the Peak, sliding towards the Trough Emergin g* 5– 20%* 2–5 years* Team 3: Cognivolv e ClaimCare – AI-Powered Claims and Authorization Data Quality Intelligence AI-ready data Past the Peak, sliding towards the Trough Emergin g* 5– 20%* 2–5 years* Team 4: Enfourcer s SDOH-Driven Member Risk and Intervention Prioritization Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years* Team 5: Innovatex CareSync – SDOH-Driven Member Risk and Intervention Prioritization Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years* Team 6: Tech Connect Provider Intelligence Engine – Network Adequacy and Access Intelligence Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years* Team 7: Tech Vanguard Meridian – AI-Powered Healthcare Claims Risk Detection and Intelligent Investigation AI agents Peak of Inflated Expectations Emergin g* 5–20% 2–5 years* Team 8: Trailblaze rs Value-Based Care Contract Performance Command Center Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years*

Team 9: Techno Squad RightPath – Avoidable Emergency Department Utilization Navigator Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years* Team 10: CodeCraft ers Claims Fraud, Waste and Abuse Risk Detector Decision intelligenc e Innovation Trigger Emergin g* 5– 20%* 2–5 years* Team 11: Fantastic Four Prior Authorization Triage and Policy Companion Generativ e AI Trough of Disillusionme nt Early mainstre am* 20– 50%* 2–5 years* Team 12: Tech Titans Prior Authorization Triage and Policy Companion AI agents Peak of Inflated Expectations Emergin g* 5–20% 2–5 years* Team 13: InsightHiv e AI-Driven Healthcare Data Quality and Anomaly Monitoring System AI-ready data Past the Peak, sliding towards the Trough Emergin g* 5– 20%* 2–5 years* Team 14: Nexora Nexora – AI-Assisted Prior Authorization Triage and Policy Companion AI agents Peak of Inflated Expectations Emergin g* 5–20% 2–5 years* * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Phase-wise Count Hype Cycle phase (2026) Projects Teams Innovation Trigger 7 Hackstreak, Enfourcers, Innovatex, Tech Connect, Trailblazers, Techno Squad, CodeCrafters Peak of Inflated Expectations 3 Tech Vanguard, Tech Titans, Nexora Past the Peak, sliding towards the Trough 3 Prime Syndicate, Cognivolve, InsightHive Trough of Disillusionment 1 Fantastic Four Total 14 Brief Notes on Each Project Team 1 – Hackstreak: Risk Adjustment and HCC Suspecting Assistant. The system predicts which members have risk conditions that may be missing from current-year documentation and ranks them for human review. Supporting a reviewer's decision with model output is what Gartner calls decision intelligence. Phase in 2026: Innovation Trigger. Team 2 – Prime Syndicate: Data Sentinel – Claims and Authorization Data-Quality Anomaly Monitor. Data Sentinel checks whether claims, pharmacy and authorization data are complete, valid

and consistent before they are used, which is the concern of Gartner's AI-ready data profile. Phase in 2026: Past the Peak, sliding towards the Trough. Team 3 – Cognivolve: ClaimCare – AI-Powered Claims and Authorization Data Quality Intelligence. ClaimCare scores data quality across six dimensions and flags anomalies before the data reaches payment and analytics systems, which is what AI-ready data is about. Phase in 2026: Past the Peak, sliding towards the Trough. Team 4 – Enfourcers: SDOH-Driven Member Risk and Intervention Prioritization. The platform ranks members by combined health and SDOH risk and recommends interventions, which is decision support of the kind Gartner calls decision intelligence. Phase in 2026: Innovation Trigger. Team 5 – Innovatex: CareSync – SDOH-Driven Member Risk and Intervention Prioritization. CareSync combines clinical and SDOH risk to tell care teams which members to support first, which falls under decision intelligence. Phase in 2026: Innovation Trigger. Team 6 – Tech Connect: Provider Intelligence Engine – Network Adequacy and Access Intelligence. The engine predicts access gaps and turns them into actions such as recruiting providers or enabling telehealth, which is decision intelligence. Phase in 2026: Innovation Trigger. Team 7 – Tech Vanguard: Meridian – AI-Powered Healthcare Claims Risk Detection and Intelligent Investigation. Meridian uses an orchestrator and specialised billing, peer and clinical agents to investigate each flagged case, which is the agentic approach Gartner placed at the Peak in 2026. Phase in 2026: Peak of Inflated Expectations. Team 8 – Trailblazers: Value-Based Care Contract Performance Command Center. The command center tells payers what changed in each ACO contract, why, and what to do next, which is decision intelligence. Phase in 2026: Innovation Trigger. Team 9 – Techno Squad: RightPath – Avoidable Emergency Department Utilization Navigator. RightPath guides members to the right care setting and shows payers where avoidable ED use occurs, which is decision support of the kind Gartner calls decision intelligence. Phase in 2026: Innovation Trigger. Team 10 – CodeCrafters: Claims Fraud, Waste and Abuse Risk Detector. The two-stage model ranks providers by fraud risk so that investigators know which case to take first, which is decision intelligence. Phase in 2026: Innovation Trigger. Team 11 – Fantastic Four: Prior Authorization Triage and Policy Companion. The system retrieves CMS policy text and uses a language model to evaluate a request and explain the recommendation, which places it under generative AI with retrieval. Phase in 2026: Trough of Disillusionment. Team 12 – Tech Titans: Prior Authorization Triage and Policy Companion. Policy criteria, clinical match and adjudication agents work in sequence on each request, which is the agentic approach Gartner placed at the Peak in 2026. Phase in 2026: Peak of Inflated Expectations. Team 13 – InsightHive: AI-Driven Healthcare Data Quality and Anomaly Monitoring System. InsightHive checks schema, completeness, validity, consistency, uniqueness and timeliness before data is used, which is the focus of AI-ready data. Phase in 2026: Past the Peak, sliding towards the Trough.

Team 14 – Nexora: Nexora – AI-Assisted Prior Authorization Triage and Policy Companion. Nexora uses Policy, Clinical Evidence, Evaluation and Critic agents to triage each request, which is the agentic approach Gartner placed at the Peak in 2026. Phase in 2026: Peak of Inflated Expectations. Meaning of the Attributes Attribute What it describes Gartner scale Hype Cycle phase Where the technology stands on the curve of expectations Innovation Trigger, Peak of Inflated Expectations, Trough of Disillusionment, Slope of Enlightenment, Plateau of Productivity Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20–50%, >50% of the target audience Time to mainstream adoption Years before the technology reaches the Plateau of Productivity <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT RISK ADJUSTMENT AND HCC SUSPECTING ASSISTANT Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 1 – Hackstreak Faculty mentors Dr. A. Ponmalar, Ms. E. Thenmozhi Students Vignesh D S – 111723201118, Thavasi Priyan P P – 111723201098, Thilak S – 111723201099, Vishaal S – 111723201111, Udaya Kumar G K – 111723201100, Sharan Babu A – 111723201089 Cognizant mentors Mr. Perumal Babu, Mr. Abdulla LNU Use case UC03 – Risk adjustment and HCC suspecting assistant Abstract Medicare Advantage insurers need a complete view of each member's chronic conditions for annual risk adjustment. Conditions documented in earlier years are sometimes missing in the current year, and some members miss care altogether. The team built a system that combines Synthea patient records, Medicare claims, prescription drug events and the ICD-10 to CMS-HCC crosswalk (about 15.9 million records), maps supported diagnoses to HCC categories and predicts next-year risk with an XGBoost model. SHAP shows the factors behind each prediction, and a locally run Qwen3 model writes a plain explanation for the human reviewer. Each case is labelled as supported, review required or a potential documentation gap; the system never invents a diagnosis or makes the final coding decision. On the test set the model reached 93.5% accuracy, F1 0.964 and ROC-AUC 0.955. Problem Statement Help Medicare Advantage insurers find members whose documented chronic conditions may not be captured in the current year, and members who have missed care, so that documentation gaps can be prioritised for human review within coding and compliance rules. Objectives • Build an integrated patient-year dataset from several Medicare sources. • Standardise ICD-10 codes and map them to CMS-HCC categories. • Predict next-year risk and flag possible documentation gaps. • Explain each flag to a human reviewer.

System Architecture Fig 1. System architecture (team's final presentation) Methodology • Data: CDC PLACES (3 million), Synthea (8.9 million), Medicare (2.3 million) and HCC crosswalk (about 45,000) records, stored in MongoDB. • Pre-processing: missing values handled, duplicates removed and ICD-10 codes standardised before mapping to HCC categories. • Features: patient-level evidence scores and the frequency and recency of each condition. • Models: Random Forest, XGBoost and LightGBM compared; XGBoost used as the main model, with a decision threshold of 0.194 chosen on validation. • Explanation: SHAP for feature importance and a local Qwen3 model for reviewer-friendly summaries. • Application: FastAPI back end with an HTML, CSS and JavaScript dashboard, ZIP upload of Synthea files and PDF reports. Technology Stack Layer Tools Machine learning Python, pandas, NumPy, XGBoost, SHAP Language model Qwen3 (local) Data MongoDB; Synthea, Medicare, CDC PLACES, HCC crosswalk Application FastAPI, HTML, CSS, JavaScript Results XGBoost evaluation on the test set of 1,505 records (team's performance slide).

Metric Value Accuracy 0.935 Precision 0.984 Recall 0.946 F1 score 0.964 ROC-AUC 0.955 PR-AUC 0.996 The minority class (no gap) had lower precision (0.52), which the team handled by tuning the decision threshold. Fig 2. Provider-level prediction dashboard (team's output screen) Outcome and SDG Mapping The approach was discussed with the Cognizant mentors on 17 August 2026, who asked for over 90% accuracy and detailed LLM explanations for each patient. The project maps to SDG 3 Good Health and Well-being. Conclusion The assistant points reviewers to documentation gaps that are supported by data, with reasons, while leaving coding decisions to people. Validation on real Medicare Advantage records would be the next step. Gartner Hype Cycle 2026: Technology Position The Risk Adjustment and HCC Suspecting Assistant project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed.

Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations* Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since a local Qwen3 model writes the reviewer explanations Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Hackstreak built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations.

Gartner Hype Cycle Graph Figure: Position of Hackstreak on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Hackstreak. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • The system predicts which members have risk conditions that may be missing from current-year documentation and ranks them for human review. Supporting a reviewer's decision with model output is what Gartner calls decision intelligence. • XGBoost reached 93.5% accuracy and an F1 score of 0.964, and SHAP shows the factors behind each flag. • The final coding decision stays with a person, which matches Gartner's view that decision intelligence augments rather than replaces human judgment.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Evidence Source code https://github.com/sharan022005/HCC Demo video https://drive.google.com/file/d/199IyhuTNrO2CeqeNWZs4y95rVUxQzYZe/view Final presentation https://drive.google.com/file/d/1humYhK60oSUAGiiZcZAi-xbXzBbLhnHV/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT DATA SENTINEL – CLAIMS AND AUTHORIZATION DATA-QUALITY ANOMALY MONITOR Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 2 – Prime Syndicate Faculty mentors Dr. N. R. Gladiss Merlin, Ms. R. Devi Students Yuvakrishnan T N – 111723201114, Shanjay K – 111723201088, Surya K – 111723201092, V. Venkata Sai Krishna – 111723201101, Vignesh M – 111723201108, Monesh S – 111723201063, Pantrangi Ruthvik – 111723201073, Mukesh Matthew Raj A – 111723201064 Cognizant mentors Mr. Saravanan Krishnappan, Mr. Tamilvanan M Use case UC10 – Claims and authorization data-quality anomaly monitor Abstract Errors in claims, pharmacy and authorization data spread to every downstream report and payment. Data Sentinel monitors these pipelines continuously. A data-quality engine checks schema, completeness, ranges, consistency and references against provider and code masters. Statistical and machine-learning detectors, namely CUSUM, the Kolmogorov–Smirnov test, median absolute deviation, Local Outlier Factor and Isolation Forest, find processing anomalies and distribution drift, and an SLA module flags pipelines likely to miss deadlines. The signals are combined into an auditable confidence score, and a retrieval-grounded Gemini layer explains the likely root cause from stage-level logs and recommends a fix. Problem Statement Detect data-quality issues, processing anomalies and SLA risk in claims, pharmacy and authorization data pipelines, and explain the likely cause with a recommended fix. Objectives • Monitor pipelines for missing, invalid, duplicate and inconsistent records. • Detect processing anomalies and distribution drift early. • Predict SLA risk before deadlines are missed. • Explain root causes and recommend remediation.

System Architecture Fig 1. System architecture (team's final presentation) Methodology • Data: beneficiary, fee-for-service and prescription drug event transactions, with authorization, provider and diagnosis and procedure code masters for reference. • Pipeline: ingestion of CSV and Parquet files, validation and cleaning, transformation with ICD- 10, HCPCS and NPI rules, and storage in DuckDB and Parquet. • Quality engine: schema, completeness, accuracy and range, consistency and referential checks, each producing pass or fail results and metrics. • Anomaly detection: CUSUM for gradual volume shifts, KS test for drift in skewed distributions, MAD as a reliable scale, LOF for local stalls and Isolation Forest for unusual records. • Evidence fusion: rule and model signals combined into a confidence score so every alert can be audited. • Root-cause analysis: RAG with Gemini restricted to pipeline telemetry, producing an explanation and recommended action. Technology Stack Layer Tools Data DuckDB, Parquet, Python Detection CUSUM, KS test, MAD, LOF, Isolation Forest Root cause RAG with Gemini Application Monitoring dashboards Results Detection methods chosen by the team and the reason for each (team's model selection appendix).

Method Used for Chosen over CUSUM Gradual shifts in claim and provider volume Static thresholds KS test Drift in right-skewed cost distributions Gaussian Z-scores MAD Robust baselines with bursty batches Standard deviation LOF Local stalls in throughput, backlog and duration Regression or PCA Grounded RAG (Gemini) Root-cause reports from logs Ungrounded LLM output Fig 2. Monitoring output screens (team's final presentation) Outcome and SDG Mapping The prototype monitors the pipeline end to end and produces root-cause reports with recommended fixes. The project maps to SDG 9 Industry, Innovation and Infrastructure and SDG 3 Good Health and Well-being. Conclusion Data Sentinel gives data-operations teams early warning of bad data and SLA risk, with explanations they can audit. Testing on live payer pipelines would confirm the alert thresholds. Gartner Hype Cycle 2026: Technology Position The Data Sentinel – Claims and Authorization Data-Quality Anomaly Monitor project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to AI-ready data. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed.

Hype Cycle attribute (2026) AI-ready data Hype Cycle phase Past the Peak, sliding towards the Trough* Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since Gemini writes the root-cause explanations Meaning of the Attributes Attribute What it describes Gartner scale AI-ready data in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Prime Syndicate worked on ai-ready data in August 2026, shortly after it had passed the Peak, when attention was moving from promise to proof of value.

Gartner Hype Cycle Graph Figure: Position of Prime Syndicate on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Prime Syndicate. It sits just past the Peak, sliding towards the Trough because ai- ready data was at that stage in 2026. At or just past the Peak, publicity and expectations are high and often run ahead of proven results. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • Data Sentinel checks whether claims, pharmacy and authorization data are complete, valid and consistent before they are used, which is the concern of Gartner's AI-ready data profile. • It combines CUSUM, the KS test, MAD, LOF and Isolation Forest so that both sudden and gradual problems are caught. • Gartner placed AI-ready data at the Peak in 2025; by 2026 attention had moved to proving its value in practice.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Slide 23 Slide 24 Evidence Source code https://github.com/Mr-Vicky-06/UC10_PipelineMonitor_DataSentinel

Demo video https://drive.google.com/file/d/1nIO-NCGtnn0gQqZEAOuTJI8ghwA32Rtl/view Final presentation https://drive.google.com/file/d/1OPpJMZRsAz1SgYa_yd5rq_pPBMfDBowu/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT CLAIMCARE – AI-POWERED CLAIMS AND AUTHORIZATION DATA QUALITY INTELLIGENCE Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 3 – Cognivolve Faculty mentors Dr. A. Ponmalar Students Swetha B – 111723201093, Renushri R – 111723201082, Darshini S – 111723201025, Roshni S S – 111723201083, Natchatra P K – 111723201067, Sadhana S – 111723201086 Cognizant mentors Mr. Saravanan Krishnappan, Mr. Tamilvanan M Use case UC10 – Claims and authorization data-quality anomaly monitor Abstract Payers process thousands of claims, pharmacy records and authorization requests daily, and missing IDs, duplicates or delayed feeds can go unnoticed until payments or reports go wrong. ClaimCare checks the data in three workflows (claims, pharmacy and authorization) for completeness, correctness, referential integrity, concordance, plausibility and internal consistency, and combines them into a data-quality score from 0 to 1. Volume behaviour is compared with each entity's historical baseline using a reliable Z-score (MAD) and the Population Stability Index, an Isolation Forest finds unusual claim batches, and Dagster tracks pipeline runs for SLA risk. SHAP explains anomalous claims, and a language model turns these findings into a business summary with a recommended action. Problem Statement Detect data-quality issues, processing anomalies and SLA risks in healthcare claims, pharmacy and authorization data early, to prevent payment errors and poor decisions. Objectives • Check incoming records against six data-quality dimensions. • Detect volume deviations and distribution shifts against historical baselines. • Predict SLA breaches from pipeline timing. • Explain anomalies and recommend corrective action.

System Architecture Fig 1. Claims workflow architecture (team's final presentation) Methodology • Data: Medicare Part D prescribers, CMS Medicare claims and CMS durable medical equipment claims. • Quality checks: completeness, correctness, referential integrity, concordance, plausibility and internal consistency, combined into a quality score (0.96 in the sample run). • Volume risk: monthly volume compared with a three-month baseline per provider, scored with a reliable Z-score (MAD) and PSI and classed as low, medium, high or critical. • Anomaly detection: Isolation Forest on claim-batch features; LOF, DBSCAN and One-Class SVM were compared. • SLA monitoring: Dagster tracks batch runs; turnaround is classed as met, minor breach or breached, and an SLA risk score from 0 to 1 is produced. • Explanation: SHAP feature contributions converted by an LLM into root cause, impact and recommended action. Technology Stack Layer Tools Data and orchestration MySQL, Python, Dagster Detection Robust Z-score (MAD), PSI, Isolation Forest, SHAP Application Flask API, monitoring dashboard Explanation LLM-generated business summaries Results Example outputs from the ClaimCare monitoring run (team's data-quality and risk slides).

Output Value Data-quality score (0–1) 0.96 SLA risk score (0–1) 0.75 (high risk) Volume risk levels Low, medium, high, critical SLA status classes Met, minor breach, breached Fig 2. ClaimCare monitoring dashboard (team's final presentation) Outcome and SDG Mapping The approach was reviewed with the Cognizant mentors in August 2026. The project maps to SDG 9 Industry, Innovation and Infrastructure and SDG 3 Good Health and Well-being. Conclusion ClaimCare stops incomplete or inconsistent data before it reaches payment and reporting systems, and explains each issue in business terms. Handling different payer data formats remains the main challenge. Gartner Hype Cycle 2026: Technology Position The ClaimCare – AI-Powered Claims and Authorization Data Quality Intelligence project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to AI-ready data. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) AI-ready data Hype Cycle phase Past the Peak, sliding towards the Trough* Maturity Emerging*

Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Decision intelligence (approaching the Peak), since SLA and volume risk scores guide operations teams Meaning of the Attributes Attribute What it describes Gartner scale AI-ready data in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Cognivolve worked on ai-ready data in August 2026, shortly after it had passed the Peak, when attention was moving from promise to proof of value.

Gartner Hype Cycle Graph Figure: Position of Cognivolve on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Cognivolve. It sits just past the Peak, sliding towards the Trough because ai-ready data was at that stage in 2026. At or just past the Peak, publicity and expectations are high and often run ahead of proven results. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • ClaimCare scores data quality across six dimensions and flags anomalies before the data reaches payment and analytics systems, which is what AI-ready data is about. • The sample run produced a data-quality score of 0.96 and an SLA risk score of 0.75, with SHAP and an LLM explaining each anomaly. • Gartner treats data readiness as a precondition for AI; this project applies it to payer data pipelines.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Slide 23 Slide 24

Slide 25 Slide 26 Slide 27 Evidence Source code https://github.com/312006Swetha/claim_care Demo video https://drive.google.com/file/d/1ne_PomPYof7fNRKPLtLCO-1bWEV5cYlr/view Final presentation https://drive.google.com/file/d/1lA6b6XXKjkrSARrLsM4PpzE8tM8ZT3JK/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT SDOH-DRIVEN MEMBER RISK AND INTERVENTION PRIORITIZATION Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 4 – Enfourcers Faculty mentors Mr. M. Stanlywit Students Swetha S – 111723201094, Priyadharshini S – 111723201077, Shalini S – 111723201087, Priyadharshini A – 111723201076, Thanuja K – 111723201096, Vasima I – 111723201104 Cognizant mentors Ms. Geetha Deepalakshmi, Mr. Vinod Vijayan Sulochana Use case UC09 – SDOH-driven member risk and intervention prioritization Abstract Most health risk models use medical and demographic data and miss social determinants of health (SDOH) such as income, housing, food and transport. The team combined 10,000 synthetic member records with county-level SDOH data from the CDC Social Vulnerability Index, CDC PLACES and USDA food access, and predicted each member's risk with a stacking ensemble of XGBoost, CatBoost and LightGBM trained with five-fold out-of-fold predictions. SHAP identifies the main risk drivers, and a FAISS-based retrieval step suggests interventions. On 2,000 test members the risk level was classified with 97.3% accuracy. A role-based dashboard with county risk maps and email alerts serves payer admins, analysts and care managers. Problem Statement Identify high-risk members early using both health and SDOH data, and support timely, equitable intervention through an analytical dashboard. Objectives • Combine member health data with county-level SDOH factors. • Predict member risk and classify it into four levels. • Explain the drivers of each member's risk. • Recommend interventions and alert care teams.

System Architecture Fig 1. Architecture diagram (team's final presentation) Methodology • Data: 10,000 synthetic members (demographics, chronic conditions, ED visits, hospitalisations, care gaps) linked by county to CDC SVI, CDC PLACES and USDA food access data. • SDOH score: the three sources combined into county-level SDOH risk factors and a risk score. • Models: XGBoost, CatBoost and LightGBM base learners combined by stacking with five-fold out-of-fold predictions; a five-model stack gave the lowest error. • Risk levels: low, medium, high and very high. • Explanation and guidance: SHAP for risk drivers; LangChain with FAISS for intervention suggestions. • Application: FastAPI and PostgreSQL back end, React and TypeScript front end with Plotly county maps, JWT authentication and role-based access. Technology Stack Layer Tools Machine learning XGBoost, CatBoost, LightGBM, stacking ensemble, SHAP Retrieval LangChain, FAISS Application FastAPI, PostgreSQL, React, TypeScript, Plotly Tools Google Colab, VS Code, GitHub Results Model comparison (risk score regression) and risk-level classification on 2,000 test members (team's performance slide).

Model MAE RMSE R² 5-model stacking 0.507 0.690 0.999 3-model stacking 0.660 0.885 0.998 CatBoost 0.667 0.906 0.998 LightGBM 1.310 1.769 0.993 XGBoost 1.508 2.069 0.990 Risk-level classification reached accuracy 0.973, precision 0.973, recall 0.973 and F1 0.973, with class F1 scores between 0.97 and 0.98. Fig 2. Role-based user interface (team's final presentation) Outcome and SDG Mapping The work was reviewed by the mentors on 15 and 19 August 2026. The project maps to SDG 3 Good Health and Well-being and SDG 10 Reduced Inequalities. Conclusion Adding SDOH factors lets the payer find vulnerable members who medical data alone would miss. The very high R² reflects synthetic data; real member data would give a more realistic estimate. Gartner Hype Cycle 2026: Technology Position The SDOH-Driven Member Risk and Intervention Prioritization project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations*

Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Composite AI, since three gradient-boosting models are stacked Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Enfourcers built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations.

Gartner Hype Cycle Graph Figure: Position of Enfourcers on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Enfourcers. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • The platform ranks members by combined health and SDOH risk and recommends interventions, which is decision support of the kind Gartner calls decision intelligence. • The five-model stacking ensemble gave the lowest error (MAE 0.507), and risk levels were classified with 97.3% accuracy. • SHAP explanations and role-based alerts help care managers act on the predictions.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Evidence Source code https://github.com/PriyadharshiniS1607/UC09-MEMBER-RISK/ Demo video https://drive.google.com/file/d/1Dp6YiRYE7DPFgvO0OX-YmjTiezBfn69e/view Final presentation https://drive.google.com/file/d/1uqM2s0dfF3QXn-I5IU7yDb3xhlp35E7B/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT CARESYNC – SDOH-DRIVEN MEMBER RISK AND INTERVENTION PRIORITIZATION Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 5 – Innovatex Faculty mentors Dr. M. Shanthi, Mr. Satheesh Abraham Leo Students Kavin Venthan Ravi Sudha – 111723201049, K Kishore – 111723201044, Varunvarshan R N – 111723201103, Vignesh S – 111723201109, Lohith S S – 111723201056, Kishore Shyam V – 111723201054, Aldrin Jhones Rishin S – 111723201005, Nithesh Kumar S – 111723201071 Cognizant mentors Ms. Geetha Deepalakshmi, Mr. Vinod Vijayan Sulochana Use case UC09 – SDOH-driven member risk and intervention prioritization Abstract Food insecurity, transport barriers and housing instability keep members from managing chronic conditions. CareSync scores each member on two fronts: a clinical risk from patient records and an SDOH risk from 12 census tract-level features, and combines them into an overall risk of low, medium or high. Patient charts uploaded as PDF or Excel are read with pypdf and GPT-4o-mini into structured data, XGBoost predicts current clinical risk and CatBoost future risk, and TreeSHAP explains each score. A LangChain agent uses the patient profile and SHAP drivers to answer questions and suggest interventions. XGBoost reached 91.6% accuracy and macro F1 0.921. Problem Statement Identify vulnerable members early by combining clinical health factors with social determinants of health, and prioritise targeted interventions for them. Objectives • Score clinical and SDOH risk separately and combine them. • Predict current and future member risk. • Explain risk with positive and negative feature contributions. • Recommend community resources and care coordination through a dashboard and agent.

System Architecture Fig 1. Architecture diagram (team's final presentation) Methodology • Community data: 12 SDOH features (poverty, housing cost burden, vehicle access, unemployment, food access and others) normalised and weighted into a community risk score. • Clinical data: patient records cleaned and engineered; PDF charts converted to structured JSON with pypdf and GPT-4o-mini. • Models: Logistic Regression, Random Forest and XGBoost compared for current risk; CatBoost with clustering for future risk. • Explanation: TreeSHAP for each member. • Agent: LangChain coordinates the models and GPT-4o-mini for summaries and recommendations. • Application: Django REST framework and PostgreSQL, React front end, Firebase authentication, Power BI and deployment on AWS. Technology Stack Layer Tools Machine learning XGBoost, CatBoost, Random Forest, Logistic Regression, TreeSHAP AI LangChain, OpenAI GPT-4o-mini Application Django, PostgreSQL, React, Firebase, Power BI Cloud AWS Results Comparison of clinical risk models (team's ML prediction slide).

Model Accuracy Macro F1 Logistic Regression 0.858 0.868 Random Forest 0.896 0.901 XGBoost 0.916 0.921 Fig 2. CareSync user interface (team's final presentation) Outcome and SDG Mapping The web application and mobile version were demonstrated at the final review in August 2026. The project maps to SDG 3 Good Health and Well-being and SDG 10 Reduced Inequalities. Conclusion CareSync links clinical risk with neighbourhood conditions so that payers can act before conditions worsen. Member-level SDOH data, rather than tract averages, would sharpen the predictions. Gartner Hype Cycle 2026: Technology Position The CareSync – SDOH-Driven Member Risk and Intervention Prioritization project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations* Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years*

Related 2026 profiles AI agents (Peak of Inflated Expectations), since a LangChain agent answers questions and suggests interventions Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Innovatex built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations.

Gartner Hype Cycle Graph Figure: Position of Innovatex on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Innovatex. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • CareSync combines clinical and SDOH risk to tell care teams which members to support first, which falls under decision intelligence. • XGBoost reached 91.6% accuracy and a macro F1 of 0.921 for clinical risk. • An agent built with LangChain and GPT-4o-mini adds a conversational layer on top of the predictions.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Slide 23 Slide 24

Slide 25 Evidence Source code https://github.com/KishoreShyam/Cognizant_Hackathon Demo video https://drive.google.com/file/d/1tFPOsu-DVIjhe7aNrbKw-rZD0NbmlDb8/view Final presentation https://drive.google.com/file/d/160bhKvPWMcDUOr2sVVHZ3468dgOLF_GZ/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT PROVIDER INTELLIGENCE ENGINE – NETWORK ADEQUACY AND ACCESS INTELLIGENCE Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 6 – Tech Connect Faculty mentors Mr. S. Nagendiran Students Ashmita B – 111723201009, Dharshine K – 111723201027, Dirisala Kavya – 111723201029, Dornala Keerthi Reddy – 111723201031, N . Pujitha Reddy – 111723201069, Kamaly B – 111723201045 Cognizant mentors Ms. Vani D, Mr. Arun Raj Dhandapani Use case UC05 – Provider network adequacy and access intelligence Abstract Payers must make sure members can reach the right provider within a reasonable distance. The Provider Intelligence Engine analyses provider supply by geography and specialty against population need and disease burden, predicts which counties have an access gap, and recommends where to add providers. County-level population, provider, utilisation and geographic data are joined on county FIPS codes and turned into features such as provider density, demand-to-supply ratio and disease burden. Logistic Regression, Decision Tree, Random Forest and XGBoost were compared for gap prediction, and a rule-based decision engine turns the risk scores into actions: recruit, refer, enable telehealth, monitor or no action. Clustering, anomaly detection, geospatial analysis, what-if simulation and alerts complete the system. Problem Statement Analyse provider availability by geography and specialty, flag likely access gaps, and recommend where the payer should add providers next. Objectives • Identify areas with insufficient coverage by specialty. • Predict county-level access gaps with machine learning. • Recommend and rank provider recruitment locations. • Simulate the effect of adding providers before contracting.

System Architecture Fig 1. Architecture diagram (team's final presentation) Methodology • Data integration: population, provider, utilisation and geographic datasets aligned by county FIPS code; disease mapped to specialty. • Features: provider density, demand–supply ratio, disease burden, utilisation, specialty availability and access-gap indicators. • Gap prediction: Logistic Regression, Decision Tree, Random Forest and XGBoost compared on accuracy, precision, recall, F1 and ROC-AUC; XGBoost probabilities passed to the decision engine. • Further modules: K-Means and DBSCAN clustering of counties, Isolation Forest for anomalous supply, Haversine-based geospatial distance, recommendation engine and Twilio-based alerts. • Decision engine: IF-THEN rules, for example high risk with low supply leads to provider recruitment and high risk with large distance leads to telehealth. • Output: interactive maps, Power BI dashboards and an LLM-powered insight assistant. Technology Stack Layer Tools Machine learning XGBoost, Random Forest, K-Means, DBSCAN, Isolation Forest Analytics Python, geospatial distance, Power BI Application FastAPI back end, React front end, Twilio alerts Results Decision rules used to convert predicted access-gap risk into actions (team's rule-based classification slide).

Condition Recommended action High risk and low supply Recruit providers High demand with a nearby alternative Refer patients High risk and large distance Enable telehealth Medium risk Monitor the area Low risk No action Fig 2. Network adequacy user interface (team's final presentation) Outcome and SDG Mapping The architecture and dashboards were refined with the Cognizant mentors on 16 and 18 August 2026. The project maps to SDG 3 Good Health and Well-being and SDG 10 Reduced Inequalities. Conclusion The engine gives network planners a ranked list of counties and specialties to target, with reasons for each recommendation. Travel-time data instead of straight-line distance would improve the access estimates. Gartner Hype Cycle 2026: Technology Position The Provider Intelligence Engine – Network Adequacy and Access Intelligence project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations*

Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Composite AI, since machine learning and a rule-based decision engine are combined Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Tech Connect built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations.

Gartner Hype Cycle Graph Figure: Position of Tech Connect on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Tech Connect. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • The engine predicts access gaps and turns them into actions such as recruiting providers or enabling telehealth, which is decision intelligence. • XGBoost probabilities feed an IF-THEN decision engine that keeps each recommendation explainable and auditable. • What-if simulation lets planners test a decision before signing provider contracts.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Slide 23 Slide 24

Slide 25 Evidence Source code https://github.com/Ashmita984/Provider-Intelligence-Engine Demo video https://drive.google.com/file/d/1blXwRRLW5bO36IFYgjqcO4fiqf7yPOfA/view Final presentation https://drive.google.com/file/d/1d5HPD6QE6lfWY5QgOQT-E6toLBX- 4PYU/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT MERIDIAN – AI-POWERED HEALTHCARE CLAIMS RISK DETECTION AND INTELLIGENT INVESTIGATION Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 7 – Tech Vanguard Faculty mentors Dr. P. Shobha Rani, Dr. B. Muthazhagan Students Abisha K M – 111723201002, S M Pooja Shree – 111723201085, N. Sri Nakshatra – 111723201117, M. Nitheesha – 111723201121, Kaviya Priya S – 111723201050, N Saranya – 111723201070, Y. Sanvi Reddy – 111723201113, Kavya S – 111723201051 Cognizant mentors Ms. Dhiravida Selvi Rajesh, Ms. Soundari Thiyagarajan Use case UC01 – Claims fraud, waste and abuse risk detector Abstract Fraud, waste and abuse in healthcare claims are hard to spot claim by claim. Meridian analyses both claim-level and provider-level behaviour with several unsupervised detectors, namely Isolation Forest, Local Outlier Factor, One-Class SVM and DBSCAN, which produce risk scores and key risk features. An LLM-powered multi-agent system then examines each case from three angles: billing patterns, comparison with peers, and clinical and rule-based checks. SHAP and generative AI give plain-language explanations, and high-risk cases are placed in an investigation queue with supporting evidence so that investigators can decide faster. Problem Statement Help payers find suspicious claims and providers among large volumes of data, explain why they were flagged, compare them with peers and prioritise them for investigation. Objectives • Detect suspicious claims and providers with complementary anomaly models. • Explain each risk score in plain language. • Benchmark providers against peers. • Prioritise high-risk cases in an investigation queue with evidence.

System Architecture Fig 1. System architecture (team's final presentation) Methodology • Data: claim-level, geography and service, and provider datasets cleaned and engineered into claim and provider features. • Detection: Isolation Forest, LOF, One-Class SVM and DBSCAN used together to score unusual claims and providers. • Multi-agent analysis: an orchestrator agent assigns billing, peer and clinical or rule-based agents, whose findings are combined into a case summary. • Explainability: SHAP for model outputs, with RAG and an LLM for narrative explanations. • Application: FastAPI back end and React investigator dashboard with overview, claims, providers and investigation queue pages. Technology Stack Layer Tools Data processing Python, pandas, SQLite Anomaly detection Isolation Forest, LOF, One-Class SVM, DBSCAN Explainable AI SHAP GenAI investigation LLM, multi-agent orchestration, RAG Application FastAPI, React Results Roles of the agents in the investigation layer (team's multi-agent architecture slide). Component Role

Anomaly models Score claims and providers and list key risk features Billing agent Reviews billing patterns behind a flag Peer agent Compares the provider with similar providers Clinical and rule agent Checks clinical plausibility and policy rules Orchestrator Combines findings and ranks the case in the queue Fig 2. Investigator dashboard (team's output demo) Outcome and SDG Mapping The team held five mentor reviews between 12 and 17 August 2026; the Cognizant mentors asked for agent evaluation, token optimisation and a fully runnable demo. The project maps to SDG 16 Peace, Justice and Strong Institutions. Conclusion Meridian turns anomaly scores into explained, prioritised investigation cases. Labelled fraud outcomes would allow the unsupervised scores to be measured and calibrated. Gartner Hype Cycle 2026: Technology Position The Meridian – AI-Powered Healthcare Claims Risk Detection and Intelligent Investigation project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to AI agents. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) AI agents Hype Cycle phase Peak of Inflated Expectations Maturity Emerging*

Market penetration 5–20% of target audience Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since LLMs write the investigation summaries Meaning of the Attributes Attribute What it describes Gartner scale AI agents in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Tech Vanguard worked with ai agents in August 2026, when it was at the Peak of the Hype Cycle. Interest was very high, while organisations were still learning where it gives dependable results.

Gartner Hype Cycle Graph Figure: Position of Tech Vanguard on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Tech Vanguard. It sits at the Peak of Inflated Expectations because ai agents was at that stage in 2026. At or just past the Peak, publicity and expectations are high and often run ahead of proven results. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • Meridian uses an orchestrator and specialised billing, peer and clinical agents to investigate each flagged case, which is the agentic approach Gartner placed at the Peak in 2026. • Isolation Forest, LOF, One-Class SVM and DBSCAN produce the risk scores the agents work from. • The agents produce evidence and a ranked queue, while the decision remains with the investigator.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Evidence Source code https://github.com/Abishakm1507/Claims-fraud-waste-and-abuse-risk-detector Demo video https://drive.google.com/file/d/13VNFU3hX2IXHJcFom4tjx4t53X2Lyd8r/view Final presentation https://drive.google.com/file/d/1FvOJPoGrCKCr5JXsdcDSOVkqV55I1u9Z/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT VALUE-BASED CARE CONTRACT PERFORMANCE COMMAND CENTER Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 8 – Trailblazers Faculty mentors Ms. C. Vimala Josphine, Mr. V. K. Sabarirajan Students Anbuaruvi R – 111723201006, Bhavya Sree P R – 111723201016, Divya K – 111723201030, Beri Sai Poojitha Reddy – 111723201015, Madhumitha S – 111723201058, Jeevalakshmi S – 111723201042 Cognizant mentors Ms. Deepika Kothandan, Mr. H G Rakesh Use case UC08 – Value-based care contract performance command center Abstract Payers managing several accountable care organisation (ACO) contracts need to know quickly what changed, why and what to do before each contract review. The command center uses CMS MSSP and provider data to segment ACOs with K-Means (silhouette 0.284), predict the next-year savings rate with Gradient Boosting, and detect unusual year-on-year changes with Isolation Forest, which outperformed LOF (F1 0.306 against 0.118). SHAP explains the drivers of risk and performance, and a local Llama 3.2 model writes a grounded contract review report from pre-computed facts, so it cannot invent numbers. All insights are shown in an interactive dashboard with alerts and an assistant. Problem Statement Give payers a single view of ACO contract performance that answers what happened, why, what will happen next and what action to take, before a contract review meeting. Objectives • Segment ACOs into performance groups. • Predict the next-year savings rate. • Detect unusual performance changes and raise alerts. • Generate grounded contract review reports.

System Architecture Fig 1. System architecture (team's final presentation) Methodology • Data: CMS MSSP ACO performance data and provider-level data, stored in Supabase (PostgreSQL); hospital and geography files excluded where no valid join existed. • Segmentation: financial, quality and utilisation indicators with median imputation and scaling; K-Means (K = 3) chosen over Agglomerative and Gaussian Mixture. • Savings rate: Gradient Boosting on previous-year savings, financial, quality, utilisation, beneficiary and provider features to predict the 2024 savings rate. • Anomalies: Isolation Forest on year-on-year change indicators, with low, medium and high severity. • Reports: Llama 3.2 (3B) run locally through Ollama on facts fetched by ACO ID and year, returned as structured JSON. • Application: FastAPI and React or Next.js dashboard with eight views including portfolio, explorer, drivers, forecasting and alerts. Technology Stack Layer Tools Machine learning K-Means, Gradient Boosting, Isolation Forest, SHAP Language model Llama 3.2 3B via Ollama (local) Data Supabase (PostgreSQL), pandas Application FastAPI, React, Next.js, Tailwind CSS, Plotly Results Model comparison for segmentation and anomaly detection (team's module slides).

Task Model Score Segmentation (silhouette) K-Means (K = 3) 0.284 Segmentation (silhouette) Agglomerative (K = 3) 0.232 Segmentation (silhouette) Gaussian Mixture (K = 3) 0.125 Anomaly detection (F1) Isolation Forest 0.306 (precision 0.657, recall 0.199) Anomaly detection (F1) Local Outlier Factor 0.118 Fig 2. Command center dashboard (team's final presentation) Outcome and SDG Mapping The full deployment was demonstrated to the mentors on 19 August 2026. The project maps to SDG 3 Good Health and Well-being. Conclusion The command center brings segmentation, forecasting, anomaly alerts and grounded reports into one view for contract reviews. The low anomaly recall shows that more labelled examples of problem ACOs are needed. Gartner Hype Cycle 2026: Technology Position The Value-Based Care Contract Performance Command Center project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations* Maturity Emerging*

Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since a local Llama 3.2 model writes the contract review report Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Trailblazers built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations.

Gartner Hype Cycle Graph Figure: Position of Trailblazers on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Trailblazers. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • The command center tells payers what changed in each ACO contract, why, and what to do next, which is decision intelligence. • K-Means segmentation (silhouette 0.284), Gradient Boosting savings forecasts and Isolation Forest alerts (F1 0.306) support the review. • The report generator receives only pre-computed facts, so the language model cannot invent figures.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Evidence Source code https://github.com/Bhavya050305/Value_based_care Demo video https://drive.google.com/file/d/1PTU6W3a0dngaqEGSLcehafptD0ehIFPF/view Final presentation https://drive.google.com/file/d/17TdfFq2ODCVogxFoDvuEB9uJuODb_x7_/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT RIGHTPATH – AVOIDABLE EMERGENCY DEPARTMENT UTILIZATION NAVIGATOR Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 9 – Techno Squad Faculty mentors Dr. G. Manikandan Students Lohitha C – 111723201057, Lakshana Sri SS – 111723201055, Keerthana Devi M – 111723201052, Namratha M – 111723201065, Anusiya S – 111723201008, Narala Dhathri – 111723201066 Cognizant mentors Mr. Durga Venkatesh Josyula, Ms. Hannah Shiny Priya Use case UC07 – Avoidable emergency department utilization navigator Abstract When members feel unwell they often cannot tell which care setting is right, so the Emergency Department (ED) becomes the default even when primary care, urgent care or telehealth would be more appropriate; payers, in turn, struggle to identify members who need proactive intervention. RightPath is a two-sided, safety-first platform. Its Member Portal offers AI-assisted symptom interaction, a deterministic safety engine for emergency warning signs, personalised care-pathway recommendations and navigation to nearby providers. Its Payer Portal offers population-level ED utilization analytics, high-utilizer identification, anomaly detection and an AI assistant. On 8,671 CMS synthetic Medicare member profiles, an XGBoost model identified high-utilization members with 93% accuracy and ROC-AUC 0.976, and an Isolation Forest flagged 174 members with unusual utilization patterns. Problem Statement Members struggle to determine the most appropriate care setting when they feel unwell, and without timely guidance the ED becomes the default option. At the payer level, repeated ED utilization makes it hard to identify members who need proactive intervention and to understand the reasons behind their utilization patterns. Key challenges: navigating between care settings, repeated high-utilization behaviour, limited visibility into unusual claims patterns, lack of transparency in AI recommendations, and linking individual care decisions with population-level financial outcomes. Objectives • Guide members safely to the most appropriate care setting for their symptoms. • Identify high-utilization and at-risk members from claims data.

• Detect unusual ED utilization patterns for review and intervention. • Give payers transparent, population-level utilization analytics and an AI assistant. System Architecture Fig 1. Architecture diagram of RightPath (team's final presentation) Methodology • Dataset: CMS synthetic Medicare data: beneficiary/enrolment, inpatient and outpatient claims: linked at member level into 8,671 member utilization profiles. • High-utilization model: XGBoost predicts high-utilization / at-risk members from engineered utilization and cost features. • Anomaly detection: Isolation Forest (300 trees, 2% contamination) flags members whose utilization differs sharply from normal behaviour. • Safety engine: deterministic rules detect emergency warning signs independently of the ML models, cost or AI responses, so emergency care is never discouraged. • Knowledge layer: retrieval-augmented generation with a Pinecone vector database and Gemini powers the Care Assistant and Payer Intelligence Assistant. • Application: Flask back end with PostgreSQL; member portal (symptoms, pathway, nearby care, history) and payer portal (members, risk, utilization, anomalies, impact). Technology Stack Layer Tools Machine learning and AI Python 3.12, XGBoost, Isolation Forest, Gemini, RAG with Pinecone Back end and data Flask, SQLAlchemy, PostgreSQL; Pytest model and safety-engine tests Front end and analytics HTML, CSS, JavaScript, Maps / Places API, Chart.js, Matplotlib, Seaborn

Results Performance of the two models (team’s performance-metrics slide). XGBoost – high-utilization pattern Value Isolation Forest – anomaly detection Value Accuracy 93% Anomalies flagged 174 Recall 89.8% Contamination 2% Precision 60.8% Trees 300 F1 score 0.725 Members analysed 8,671 ROC-AUC 0.976 Learning type Unsupervised Fig 2. Model performance and ED utilization analysis (team's final presentation) Outcome and SDG Mapping A working two-portal prototype was developed and presented at the final review in August 2026. The project maps to SDG 3 Good Health and Well-being. Conclusion RightPath connects individual care decisions with population-level outcomes: members get safe, explainable guidance to the right care setting, and payers see where avoidable ED use and unusual patterns occur. Future work includes validation on real payer data, real-time claims integration and measurement of avoided ED visits. Gartner Hype Cycle 2026: Technology Position The RightPath – Avoidable Emergency Department Utilization Navigator project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence.

Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations* Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since Gemini powers the care and payer assistants Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Techno Squad built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations.

Gartner Hype Cycle Graph Figure: Position of Techno Squad on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Techno Squad. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • RightPath guides members to the right care setting and shows payers where avoidable ED use occurs, which is decision support of the kind Gartner calls decision intelligence. • XGBoost identified high-utilization members with 93% accuracy and ROC-AUC 0.976, and Isolation Forest flagged 174 unusual members. • A deterministic safety engine makes sure emergency care is never discouraged by the models.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15

Evidence Source code https://github.com/NAMRATHAMURUGAN/avoidable-ed-utilization-navigator- app Demo video https://drive.google.com/file/d/1_pReWmbhdrYZpm8_3h2f5Df7qtjyn34a/view Final presentation https://drive.google.com/file/d/15IUJTkBxz7JaTyR-dvhcouhNebDnoekX/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT CLAIMS FRAUD, WASTE AND ABUSE RISK DETECTOR Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 10 – CodeCrafters Faculty mentors Dr. P. Shobha Rani Students Kishor S – 111723201053, Kathirvel M – 111723201047, Harshwanth Oduru – 111723201033, Karumuri Thanmai – 111723201046, Billgates E – 111723201018, P.Gowtham Reddy – 111723201074 Cognizant mentors Ms. Soundari Thiyagarajan, Ms. Dhiravida Selvi Rajesh Use case UC01 – Claims fraud, waste and abuse risk detector Abstract Fraud often looks like normal billing when claims are checked one at a time, and investigators cannot review every provider. The team built a two-stage provider risk detector on the Kaggle Medicare provider fraud data enriched with CMS program integrity guidance. A Random Forest first screens providers using out-of-bag probabilities, and XGBoost then scores the suspicious ones in detail. Investigators can drill down from a provider score to the claims that triggered suspicious rules, see SHAP factors, and read an explanation produced by RAG over CMS policy and an LLM served through Groq. An investigator chatbot and a human decision step (confirm fraud, clear or request information) complete the workflow. At a threshold of 0.382 the system detected 388 of 506 fraudulent providers. Problem Statement Find suspicious providers among large numbers of claims, detect unusual billing and hidden provider–physician–beneficiary links, explain why a provider is risky and help investigators decide which case to take first. Objectives • Screen providers for fraud risk at scale. • Score suspicious providers in detail. • Explain each score with SHAP and policy context. • Support investigator decisions with drill-down and a chatbot.

System Architecture Fig 1. Architecture diagram (team's final presentation) Methodology • Data: Kaggle healthcare provider fraud dataset (inpatient, outpatient, beneficiary and labels) and CMS program integrity and NCCI manuals; 5,410 providers with a 9.4% fraud rate. • Features: provider-level billing, reimbursement, deductible, claim count and beneficiary features; collinear billing features examined. • Stage 1: Random Forest with balanced class weights screens providers using out-of-bag scores. • Stage 2: XGBoost scores the surviving providers, with the threshold tuned for recall. • Explanation: SHAP plus retrieval over CMS policy (ChromaDB and reciprocal rank fusion) and an LLM via Groq. • Application: FastAPI back end, React and TypeScript front end, deployed on Railway and Vercel with Docker. Technology Stack Layer Tools Machine learning Random Forest, XGBoost, SHAP Retrieval and LLM ChromaDB, reciprocal rank fusion, Groq Application FastAPI, React, TypeScript, Tailwind CSS Deployment Docker, Railway, Vercel, GitHub Results Confusion matrix of the two-stage model at a threshold of 0.382 (team's visualisation slide). Predicted not fraud Predicted fraud

Actual not fraud (4,904) 4,657 247 Actual fraud (506) 118 388 This gives a recall of about 0.77 and a precision of about 0.61 for the fraud class. Fig 2. Investigator user interface (team's final presentation) Outcome and SDG Mapping The prototype was demonstrated to the Cognizant mentors on 18 and 19 August 2026 and deployed online. The project maps to SDG 16 Peace, Justice and Strong Institutions. Conclusion The two-stage design narrows thousands of providers to a ranked, explained shortlist for investigators. Reducing false positives (247 providers) is the main area for improvement. Gartner Hype Cycle 2026: Technology Position The Claims Fraud, Waste and Abuse Risk Detector project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Decision intelligence. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Decision intelligence Hype Cycle phase Innovation Trigger, approaching the Peak of Inflated Expectations* Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since an LLM explains each provider's risk from CMS policy

Meaning of the Attributes Attribute What it describes Gartner scale Decision intelligence in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. CodeCrafters built this system in August 2026, when decision intelligence was still approaching the Peak of the Hype Cycle and used by a minority of organisations. Gartner Hype Cycle Graph Figure: Position of CodeCrafters on the Gartner Hype Cycle for AI, 2026

Reading the Graph The red dot marks CodeCrafters. It sits near the end of the Innovation Trigger, approaching the Peak because decision intelligence was at that stage in 2026. In this phase a technology attracts growing attention, but adoption is still limited. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • The two-stage model ranks providers by fraud risk so that investigators know which case to take first, which is decision intelligence. • At a threshold of 0.382 the system detected 388 of 506 fraudulent providers. • SHAP, policy retrieval and a human decision step (confirm, clear or request information) keep the investigator in control.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Evidence Source code https://github.com/kishor-santhoshkumar/Claims-Fraud-Risk-Detector Application https://frontend-theta-eight-42.vercel.app/ Demo video https://drive.google.com/file/d/14Bj9X_G_FP183PVtaf5uc9kINOlKyWRm/view Final presentation https://drive.google.com/file/d/1kqn5uMisd9PXxbqPj8ym3PGFCQtJzRfU/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT PRIOR AUTHORIZATION TRIAGE AND POLICY COMPANION Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 11 – Fantastic Four Faculty mentors Dr. R. Priyanka Pramila Students Ajey K – 111723201003, Jaikishore P – 111723201038, Ajey Adithyan A – 111723201001, Chaluvadi Bramhendra Kumar – 111723201021, Valikela Surya Rajendra – 111723201120, Jaganathan V – 111723201037, Cheran M – 111723201301, Bathala Maveen – 111723201013 Cognizant mentors Mr. Manikandaraj, Mr. Dhruv Kumar Use case UC02 – Prior authorization triage and policy companion Abstract Prior authorization is slow and inconsistent because reviewers must match each request to complex Medicare coverage rules. The team built a system that extracts patient, diagnosis (ICD-10) and procedure (CPT/HCPCS) details from a request, finds the relevant National Coverage Determination, Local Coverage Determination or Article through the CMS Coverage API, and evaluates the request against the policy text using retrieval-augmented generation. Deterministic article and administrative checks validate the codes, and a decision engine recommends approve, deny or pend for nurse review. A language model then writes a plain explanation listing the evidence, criteria met or not met, and any missing information. Results are shown in a React portal and a Power BI dashboard. Problem Statement Take an incoming prior authorization request, extract the key clinical and administrative facts, check them against the applicable coverage policy, and recommend approve, deny or pend for nurse review with reasons. Objectives • Convert a PA request into structured data. • Identify the applicable NCD, LCD or Article. • Evaluate the request against policy criteria with RAG and rules. • Explain the recommendation to the reviewer.

System Architecture Fig 1. System architecture (team's final presentation) Methodology • Phase 1–2: request data extraction and CMS policy identification using ICD-10 and CPT/HCPCS codes through the CMS Coverage API. • Phase 3–5: policy routing to NCD, LCD or Article, with policy documents embedded for semantic search and comparison against the request. • Phase 6: deterministic article and administrative validation of codes and requirements. • Phase 7: decision engine recommending approve, deny or pend for nurse review. • Phase 8: LLM synthesis of decision summary, policy evidence, criteria status and missing information. • Application: FastAPI back end, MongoDB, React front end and Power BI dashboard. Technology Stack Layer Tools Policy data CMS Coverage API (NCD, LCD, Articles), ICD-10, CPT/HCPCS AI Embeddings and vector search (RAG), LLM summarisation, NLP extraction Application FastAPI, Python, MongoDB, React Reporting Power BI Results Evaluation measures and target levels set by the team for the RAG component (team's performance evaluation slide). Measure Target

Policy retrieval precision 90% Recall 85% Mean reciprocal rank 90% Faithfulness / groundedness 90% Answer relevance 90% These are target levels; measured values were not reported in the final presentation. Fig 2. System demo screens (team's final presentation) Outcome and SDG Mapping The prototype was presented at the final review in August 2026. The project maps to SDG 3 Good Health and Well-being. Conclusion The system links every recommendation to the CMS policy that supports it and sends uncertain cases to a nurse. Measuring the RAG targets on a labelled set of PA cases is the next step. Gartner Hype Cycle 2026: Technology Position The Prior Authorization Triage and Policy Companion project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to Generative AI. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) Generative AI Hype Cycle phase Trough of Disillusionment Maturity Early mainstream* Market penetration 20–50% of target audience*

Time to mainstream adoption 2–5 years* Related 2026 profiles AI agents (Peak of Inflated Expectations), relevant if the phases are turned into separate agents Meaning of the Attributes Attribute What it describes Gartner scale Generative AI in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Early mainstream. The technology is proven, products are widely available and adoption is growing quickly. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 20–50%. Between a fifth and a half of target organisations already use it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Fantastic Four used generative ai in August 2026, when it was in the Trough of Disillusionment and the focus had moved to reliability, grounding and return on investment.

Gartner Hype Cycle Graph Figure: Position of Fantastic Four on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Fantastic Four. It sits in the Trough of Disillusionment because generative ai was at that stage in 2026. In the Trough interest falls, and only the uses that give clear results continue. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • The system retrieves CMS policy text and uses a language model to evaluate a request and explain the recommendation, which places it under generative AI with retrieval. • Deterministic code and administrative checks sit alongside the model, and uncertain cases are sent for nurse review. • Gartner placed generative AI in the Trough in 2026, as organisations focused on reliability and grounding, which this design addresses.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Evidence Source code https://github.com/1038-Jaikishore/prior-authorisation/tree/main/cms-prior-auth Demo video https://drive.google.com/file/d/1ktV7dLonnphKcIHHe3Xcgb850_xogF4j/view Final presentation https://drive.google.com/file/d/1LYBzCxbz8gMfE3S0zQGISkZJDXD2QmXP/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT PRIOR AUTHORIZATION TRIAGE AND POLICY COMPANION Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 12 – Tech Titans Faculty mentors Dr. S. Srijayanthi, Ms. B. Mythili Students Akash E – 111723201004, Austin Joe T – 111723201011, Chaluvadi Sai Nivas – 111723201022, Dineshkumar C – 111723201028, Tharun Rex T – 111723201097, S.harish – 111723201084, Barath Kumar S – 111723201012, Bavanesh P – 111723201014 Cognizant mentors Ms. Bhuvaneshwari Thangaraj, Mr. Srihari Jaivankumar Use case UC02 – Prior authorization triage and policy companion Abstract The team built a policy-grounded prior authorization system that links each request to the right CMS policy, the relevant patient evidence and a traceable recommendation. A provider portal captures patient, coverage, coding and clinical evidence. The request passes through a structured CMS policy lookup, BGE-M3 hybrid retrieval over policy text stored in PostgreSQL with pgvector, evidence assembly, and three agents for policy criteria, clinical matching and adjudication, followed by groundedness and safety checks. The recommendation is approve, request more information or pend; final approval or denial stays with the human reviewer, who works in a reviewer portal with an evidence matrix, citations, RFI workflow and audit trace. Problem Statement Provide an intelligent, policy-grounded system that makes prior authorization faster, more consistent and easier to get right, without replacing human judgment. Objectives • Connect each PA request to the applicable CMS policy. • Match patient evidence to policy criteria. • Recommend approve, request more information or pend with grounded reasons. • Support reviewers with evidence, citations and an audit trail.

System Architecture Fig 1. Architecture diagram (team's final presentation) Methodology • Clinical evidence and data: clinical notes processed, linked to criteria and checked for traceability and quality. • Policy and RAG: CMS policy documents structured and indexed with BGE-M3 hybrid retrieval in PostgreSQL with pgvector. • Agentic workflow: policy criteria, clinical match and adjudication agents built with LangGraph, with groundedness and safety checks. • Back end: FastAPI services for submission, retrieval, case management and auditing. • Front end: React and Vite provider portal and reviewer portal. • Infrastructure: Docker Compose, environment-based configuration and fail-fast handling of Groq rate limits. Technology Stack Layer Tools AI LangGraph agents, BGE-M3 hybrid retrieval, Groq LLM Data PostgreSQL, pgvector Application FastAPI, React, Vite Infrastructure Docker Compose Results Recommendation outcomes produced by the adjudication step (team's solution slide). Outcome When it is given

Approve All mandatory criteria satisfied with grounded evidence Request more information Required evidence missing or incomplete Pend Evidence conflicts, criteria fail, policy cannot be established or an agent is unavailable Fig 2. Provider portal (team's final presentation) Outcome and SDG Mapping The provider and reviewer portals were demonstrated at the final review in August 2026. The project maps to SDG 3 Good Health and Well-being. Conclusion The design keeps every AI recommendation traceable to policy text and patient evidence, with humans making the final decision. Measured accuracy on labelled PA cases would strengthen the evaluation. Gartner Hype Cycle 2026: Technology Position The Prior Authorization Triage and Policy Companion project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to AI agents. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) AI agents Hype Cycle phase Peak of Inflated Expectations Maturity Emerging* Market penetration 5–20% of target audience Time to mainstream adoption 2–5 years*

Related 2026 profiles Generative AI (Trough of Disillusionment), since LLMs read clinical evidence and policy text Meaning of the Attributes Attribute What it describes Gartner scale AI agents in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Tech Titans worked with ai agents in August 2026, when it was at the Peak of the Hype Cycle. Interest was very high, while organisations were still learning where it gives dependable results.

Gartner Hype Cycle Graph Figure: Position of Tech Titans on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Tech Titans. It sits at the Peak of Inflated Expectations because ai agents was at that stage in 2026. At or just past the Peak, publicity and expectations are high and often run ahead of proven results. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • Policy criteria, clinical match and adjudication agents work in sequence on each request, which is the agentic approach Gartner placed at the Peak in 2026. • BGE-M3 hybrid retrieval grounds the agents in CMS policy text, and groundedness and safety checks run before any recommendation. • Final approval or denial stays with the human reviewer, as Gartner recommends for agentic systems in regulated work.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Evidence Demo video https://drive.google.com/file/d/1-O5zUJ7FvMYgckkq-oCP633nzf0Fca_W/view Final presentation https://drive.google.com/file/d/1KQDh5WgiXfBs8956Ws9OqN1YK5FCY_vI/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT AI-DRIVEN HEALTHCARE DATA QUALITY AND ANOMALY MONITORING SYSTEM Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 13 – InsightHive Faculty mentors Ms. C. Vimala Josphine, Dr. A. Vijayaraj Students Dhanush R Y – 111723201026, Janarthanan V – 111723201039, Darshan Sanjay D – 111723201024, Jency Jovita SM – 111723201043, Samyuktha RS – 111723201122, Ramya P – 111723201080, H Praveen Kumar – 111723201034, Dandu Umesh Kummar Reddy – 111723201023 Cognizant mentors Mr. M. Tamilvanan, Mr. Saravanan Krishnappan Use case UC10 – Claims and authorization data-quality anomaly monitor Abstract InsightHive monitors claims, pharmacy and authorization data for quality problems, processing anomalies and SLA risk. On a dataset of 10,000 records with 32 fields, the system checks schema, completeness, validity, consistency, uniqueness and timeliness. An anomaly engine combines statistical methods (Z-score, IQR), correlation analysis and an Isolation Forest, which the team chose after comparing it with LOF and One-Class SVM. SLA monitoring covers time-based, service-based and pending SLAs, and a weighted anomaly score ranks issues by severity. A RAG and root-cause- analysis agent using ChromaDB and a language model explains likely causes and recommends fixes in a dashboard. Problem Statement Detect data-quality issues, processing anomalies and SLA risks in claims, pharmacy or authorization pipelines, explain the likely causes and recommend fixes. Objectives • Check data quality automatically across six dimensions. • Detect anomalies and SLA breaches. • Score and rank issues by severity. • Explain root causes and recommend corrective action.

System Architecture Fig 1. Architecture diagram (team's final presentation) Methodology • Data: 10,000 records with 32 fields covering provider, service, claim, financial, authorization and SLA information. • Quality checks: schema, completeness, validity, consistency, uniqueness and timeliness. • Feature engineering: cleaning, transformation and derived features for anomaly detection. • Anomaly engine: Z-score and IQR, correlation analysis and Isolation Forest, compared with LOF and One-Class SVM. • SLA monitoring and scoring: time-based, service-based and pending SLA checks; composite weighted score, severity levels and entity-level scores. • Explanation: RAG with ChromaDB and an RCA agent with an LLM; dashboard front end. Technology Stack Layer Tools Detection Python, Isolation Forest, Z-score, IQR Explanation RAG, ChromaDB, LLM-based RCA agent Application Back-end services and monitoring dashboard Results Comparison of anomaly detection methods (team's Isolation Forest vs LOF vs One-Class SVM slide). Aspect Isolation Forest LOF One-Class SVM Approach Isolates points with random splits Local density Learned boundary

Suited to large data Yes Limited Limited Selected Yes No No Fig 2. Anomaly detection results (team's final presentation) Outcome and SDG Mapping The workflow and architecture were reviewed with the mentors on 15 August 2026. The project maps to SDG 9 Industry, Innovation and Infrastructure and SDG 3 Good Health and Well-being. Conclusion InsightHive turns pipeline checks into a ranked list of explained issues. Measuring detection accuracy on labelled incidents would show how well the weighting works. Gartner Hype Cycle 2026: Technology Position The AI-Driven Healthcare Data Quality and Anomaly Monitoring System project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to AI-ready data. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) AI-ready data Hype Cycle phase Past the Peak, sliding towards the Trough* Maturity Emerging* Market penetration 5–20% of target audience* Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since an RCA agent explains issues with an LLM

Meaning of the Attributes Attribute What it describes Gartner scale AI-ready data in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. InsightHive worked on ai-ready data in August 2026, shortly after it had passed the Peak, when attention was moving from promise to proof of value. Gartner Hype Cycle Graph Figure: Position of InsightHive on the Gartner Hype Cycle for AI, 2026

Reading the Graph The red dot marks InsightHive. It sits just past the Peak, sliding towards the Trough because ai-ready data was at that stage in 2026. At or just past the Peak, publicity and expectations are high and often run ahead of proven results. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • InsightHive checks schema, completeness, validity, consistency, uniqueness and timeliness before data is used, which is the focus of AI-ready data. • Isolation Forest was chosen over LOF and One-Class SVM for large datasets, and a weighted score ranks issues by severity. • A RAG-based root-cause agent explains each problem and suggests a fix.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Slide 23 Slide 24 Evidence Source code https://github.com/Ramya-Reshma/CTS-Hackathon

Demo video https://drive.google.com/file/d/1UaigKbotoGt4_rfhm2wnRJolK-HVjA4y/view Final presentation https://drive.google.com/file/d/1x2U2RgfN4QipHmtYz-xwFSf0yZQpDtfj/view

R.M.K. ENGINEERING COLLEGE (An Autonomous Institution) R.S.M NAGAR, KAVARAIPETTAI – 601 206. Gummidipoondi Taluk, Thiruvallur District, Tamil Nadu. Department of Artificial Intelligence and Data Science PROJECT REPORT NEXORA – AI-ASSISTED PRIOR AUTHORIZATION TRIAGE AND POLICY COMPANION Cognizant Hack to Hire 2027 · Academic Year 2026-2027 Team Team 14 – Nexora Faculty mentors Dr. R. Priyanka Pramila Students Vedarathna A K – 111723201107, Mittapalli Lokeswar Reddy – 111723201061, Thangavelu Adithya – 111723201304, Shaik Aaqil – 111723201123, Darsi Chenchu Manideep – 111723201116, R Nikhil – 111723201078, Naveen Krishnan R – 111723201068 Cognizant mentors Ms. Bhuvaneshwari Thangaraj, Mr. Srihari Jaivankumar Use case UC02 – Prior authorization triage and policy companion Abstract Prior authorization takes 3 to 14 days and depends on manual review of clinical notes against layered Medicare rules. Nexora is a deterministic hybrid system that automates the first triage of a request while keeping the recommendation evidence-based and reviewable. It extracts procedure, diagnosis, jurisdiction and clinical information, resolves the policy chain from National Coverage Determination through jurisdiction to Local Coverage Determination and Article, and evaluates the request twice: a structured SQL check of HCPCS/CPT and ICD-10 codes, and a semantic multi-agent analysis of clinical notes using RAG over pgvector. A Critic Agent checks every AI citation against the raw notes, and SQL rules override the AI where they apply. The output is approve, pend for nurse review, request more information or reject, with reasons. The application is deployed on AWS with Bedrock models and was tested with more than 230 test cases. Problem Statement Automate the initial triage of prior authorization requests by combining clinical evidence, administrative facts and coverage policies, while keeping every decision transparent and reviewable. Objectives • Extract and normalise request details. • Find the applicable policy chain and retrieve policy evidence. • Evaluate requests with deterministic rules and agentic clinical analysis. • Produce an explainable recommendation for the reviewer.

System Architecture Fig 1. System architecture (team's final presentation) Methodology • Data: Medicare NCDs, LCDs and Articles from the CMS coverage database, ICD-10-CM and CPT/HCPCS code tables, and Synthea patient records with physician notes and therapy history; all de-identified. • Storage: PostgreSQL for codes, policy criteria and requests; pgvector with HNSW indexing for policy embeddings. • Agents: Policy Agent, Clinical Evidence Agent, Evaluation Agent and Critic Agent orchestrated over the RAG pipeline. • Authority ladder: SQL pre-filtering of hard code matches and exclusions before any LLM call; SQL overrides AI where both apply. • Optimisation: prompt context trimmed by over 60%, incomplete requests flagged before LLM use, and inference moved from local LM Studio to AWS Bedrock. • Testing and deployment: 230+ test cases; deployed on AWS (CloudFront) with Qwen models. Technology Stack Layer Tools AI Multi-agent pipeline, RAG, Qwen models on AWS Bedrock Data PostgreSQL, pgvector, CMS coverage data, Synthea Application FastAPI, React Cloud AWS (Bedrock, CloudFront) Results Engineering outcomes reported by the team (team's system and performance optimisation slide).

Item Result Prompt context reduction Over 60% Test cases defined 230+ Inference latency From minutes (local) to sub-second (AWS Bedrock) Decision outcomes Approve, pend, request more information, reject Fig 2. Request input and output format (team's final presentation) Outcome and SDG Mapping The mentors reviewed the work on 17 and 18 August 2026, asking for an impact section, handling of rejected cases and better cost efficiency, which the team added. The project maps to SDG 3 Good Health and Well-being. Conclusion Nexora shows how deterministic rules and AI agents can share prior authorization triage safely, with the rules taking precedence. Accuracy on a larger set of real PA cases would be the next measure. Gartner Hype Cycle 2026: Technology Position The Nexora – AI-Assisted Prior Authorization Triage and Policy Companion project was done in August 2026 under the Cognizant Hackathon for the 2027 batch (Hack to Hire 2027). Its main technology has been compared with the Gartner Hype Cycle for Artificial Intelligence, 2026, the edition for the year in which the project was done. On that Hype Cycle the project comes closest to AI agents. Gartner's benefit rating is left out, since the use case came from Cognizant and the business value of the solution was not assessed. Hype Cycle attribute (2026) AI agents Hype Cycle phase Peak of Inflated Expectations Maturity Emerging*

Market penetration 5–20% of target audience Time to mainstream adoption 2–5 years* Related 2026 profiles Generative AI (Trough of Disillusionment), since Qwen models on AWS Bedrock generate the evidence analysis Meaning of the Attributes Attribute What it describes Gartner scale AI agents in 2026 Maturity How far the technology itself has developed Embryonic, Emerging, Adolescent, Early mainstream, Mature mainstream, Legacy Emerging. First commercial products are available, but use is mostly limited to pilots and early adopters, and solutions still need considerable customisation. Market penetration The share of the intended users who already use it <1%, 1–5%, 5–20%, 20– 50%, >50% of the target audience 5–20%. Roughly one in ten target organisations had adopted it. Time to mainstream adoption The number of years before it reaches the Plateau of Productivity, where it is widely used and gives dependable results <2, 2–5, 5–10, >10 years, or obsolete before reaching the plateau 2–5 years. Routine use was expected between about 2028 and 2031. * Gartner's Hype Cycle for Artificial Intelligence, 2026 was published on 28 August 2026 and its profile figures are not public. Values marked * are carried forward from Gartner's 2024 and 2025 profiles or assigned on Gartner's scale from the 2026 position. The AI agents penetration (5–20%) reflects Gartner's 2026 finding that 17% of organisations had deployed agents. Nexora worked with ai agents in August 2026, when it was at the Peak of the Hype Cycle. Interest was very high, while organisations were still learning where it gives dependable results.

Gartner Hype Cycle Graph Figure: Position of Nexora on the Gartner Hype Cycle for AI, 2026 Reading the Graph The red dot marks Nexora. It sits at the Peak of Inflated Expectations because ai agents was at that stage in 2026. At or just past the Peak, publicity and expectations are high and often run ahead of proven results. The grey dots are other AI technologies from the same Hype Cycle, shown for comparison. How the Project Relates to the Profile • Nexora uses Policy, Clinical Evidence, Evaluation and Critic agents to triage each request, which is the agentic approach Gartner placed at the Peak in 2026. • A deterministic SQL authority ladder overrides the AI where rules apply, and the Critic Agent checks every citation against the raw notes. • Prompt context was cut by over 60% and the system was tested with more than 230 cases.

Annexure: Final Project Presentation Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8

Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16

Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Evidence Source code https://github.com/Veda-rathna/prior-auth-policy-companion Application https://d13yiinz4ntt17.cloudfront.net/ Demo video https://drive.google.com/file/d/1VRBoCHqtVMUBYg8yEPkKdwf9Ex83aL8V/view Final presentation https://drive.google.com/file/d/1ZxV4tClezqqzgYeCTIq-A1HtvNCpxicc/view