LLM-driven SQL Generation and Analysis System using LangGraph, FastAPI, PostgreSQL, FAISS • Implemented pipelines for a system that converts natural language queries into executable SQL over multi-table PostgreSQL datasets, supporting complex analytical queries such as aggregations, joins, and ranking. • Coordinated agent-based workflows using LangGraph, enabling iterative query refinement, conditional execution, and failure recovery, reducing invalid query generation by 40%. • Devised a lightweight tool server (MCP-style API) using FastAPI to expose database operations for improved modularity. • Integrated a retrieval layer using FAISS and sentence-transformer embeddings to store and reuse past successful queries, improving SQL generation accuracy over time. • Added a human feedback loop to capture correctness signals, storing successful queries and tracking failed patterns to guide future work. • Enforced schema-aware SQL generation with strict constraints and encouraged use of window functions for efficient analytical queries