CodeRAG Agent

Technical Description

• Engineered an AI-powered code analysis system using Tree-sitter to generate AST-aware code chunks and comprehensive dependency graphs, successfully processing codebases with 10,000+ lines across multiple programming languages.
• Designed and deployed a multi-language repository ingestion pipeline that extracts function definitions, resolves call hierarchies, maps import dependencies, and persists structured relationships in Neo4j graph database for efficient querying and traversal.
• Implemented an agentic workflow with LangChain and LangGraph featuring context retrieval tools, top-k vector similarity search (cosine similarity) using Pinecone, and conversational memory, achieving sub-3-second response times with 90%+ answer accuracy and reduced LLM hallucinations.
• Developed an interactive React frontend with real-time search capabilities, syntax-highlighted code visualization, and semantic relationship mapping that displays function dependencies and call hierarchies in an intuitive interface.
• Integrated GitHub API for seamless repository cloning and FastAPI backend for efficient processing, enabling developers to analyze any public GitHub repository through natural language queries without manual code inspection.

Tech Stack