RAG PIPELINES
Developed and evaluated a 7-agent LangGraph-based AI system for real-time financial news intelligence. Processed and deduplicated 500+ articles per day with 97% accuracy using semantic similarity + metadata matching. Implemented and optimized RAG pipelines (using FAISS, semantic chunking, hybrid search) to ground LLM responses in live news data. Performed extensive model evaluation including: - Hallucination reduction (32% improvement vs baseline) - Entity-level precision and recall metrics - LLM-as-a-Judge scoring to compare agent behaviors and output quality - User satisfaction evaluation (92% on generated Python trading strategies) Deployed production-grade system with JWT authentication, encryption, and audit logging using FastAPI and Docker.