claims-analysis

Python toolkit for Medicaid claims data analysis — preprocessing, cleaning, risk adjustment (Elixhauser, CDPS-Rx), quality measures (PQI, BETOS, low-value care), and patient-level analytic file construction for MAX and TAF CMS data. Built on Dask for scalable processing of large-scale healthcare claims datasets.

AI-powered VA claims analysis — condition discovery, decision review, and evidence-backed appeal preparation

End-to-end insurance portfolio analytics project using SQL and Power BI to analyse profitability, claims risk, exposure and management actions.

Insurance Claims Analytics Dashboard | Power BI | DAX | SQL | Data Visualization

Power BI dashboard analyzing healthcare consumption for a TPA company, with insights on utilization, costs, members, and providers.

Experience reporting and monitoring on tidy data: summaries, actual-vs-expected, claimant concentration, decomposition, one-call Excel packets.

SQL analytics on a CMS-style healthcare data warehouse — PMPM metrics, provider outlier detection, enrollment trends, and program outcomes across 5 analytical sections

SQL analysis of healthcare claims denials, AR aging and appeal recovery using MySQL

Power BI dashboard for insurance risk and claims analysis, highlighting policy distribution, claim exposure, and segment-wise business insights.

Medicaid intelligence workspace for claims analytics, risk scoring, and population-level healthcare insights.

End-to-end healthcare analytics using PostgreSQL, SQL, and Python. Includes data cleaning, ERD, 20 business KPIs, automated query execution, and CSV exports for patient, payer, and physician insights.

End-to-end Healthcare Data analytics project analyzing 3.15 GB of CMS claims data using Alteryx, SQL Server, and Power BI.

Power BI dashboard for PRISM Insurance — tracks ₹5.98M premium, ₹600.55M coverage & ₹16.91M claims across Auto, Health, Life, Travel and Home policies

Autonomous AI agent that tests hypotheses against a 10K-record synthetic healthcare claims dataset and detects/corrects its own analytical errors — 5/5 hypotheses validated, self-correction resolved on 4/5 across four injected failure modes.