Apaar
Saroj

Healthcare systems worldwide face a fundamental question: are they getting enough health benefit per dollar spent? This project builds a composite efficiency metric — the Healthcare Efficiency Score (HES) — using objective entropy-based weighting so the data decides what matters, not arbitrary assumptions.

Causal inference via synthetic control or DiD. Time-lagged spending features. Sub-national data extension. Uncertainty bands on the policy simulator.

Wind turbine failures are costly and hard to predict. Unplanned downtime can run into hundreds of thousands of dollars per turbine per day. The challenge was building a system that ingests operational and weather data, detects early failure signals, and surfaces alerts before the turbine goes down.

Credit risk analysts need to quickly identify which borrower segments carry the highest default risk — and why. With 14.5M LendingClub loan records spanning 2007–2018, this project builds an interactive 3-tab Tableau dashboard that lets analysts explore default patterns, rejection drivers, and geographic risk without writing a single query.

Real-world AI systems often need to combine models trained on incompatible label systems. This project explores whether a small calibration set can learn that mapping automatically — rather than hardcoding it by hand.

Replace greedy mapping with optimal transport. Add a zero-shot LLM baseline. Use LoRA instead of full fine-tuning. Generate a non-synthetic ground truth dataset.