Projects

CFO-endorsed Data Analyst AI and RAG Agent built at Kaiser Permanente to upload data, query in natural language, and generate automated, evidence-based reports. Reduced time to analysis from hours to <2 minutes>, scaled to analyze 10+ Excel pages in real-time.

• Engineered evaluation pipeline reducing hallucination rates by 60%+
• Full-stack delivery: Python/ML backend + UI
• Random Forest models for hospital bottleneck prediction

Engineered a RAG-based patient portal (Next.js, Python) that analyzes user symptoms and allows providers to instantly query medical records, significantly reducing chart review time.

Engineered a multi-agent legal platform (FastAPI, Next.js) that automates intake via RAG-based case analysis, generating a weighted "Case Strength Score" (Based on Liability, Evidence and Damages) to prioritize high-value claims.

Implemented a Deep Q-Learning agent using PyTorch that learns to play Snake through reinforcement learning. The agent uses an 11-state input (danger detection, direction, food location) and a 256-node hidden layer neural network to predict optimal moves.

• Experience replay with 100K memory buffer for stable learning
• Epsilon-greedy exploration strategy balancing exploration/exploitation
• Bellman equation implementation for Q-value updates (γ = 0.9)