GitDecode Vectorless RAG & Graph-Native AST Engine
Architected a Vectorless RAG & Graph-Native AST Engine using FastAPI and Supabase to parse codebases into AST knowledge graphs for precise LLM context-stuffing.
Project Case Study
GitDecode is an AI-powered codebase analysis platform utilizing a Vectorless RAG & Graph-Native AST Engine to parse codebases into AST knowledge graphs.
Key Highlights: - Architected a Vectorless RAG & Graph-Native AST Engine using FastAPI and Supabase to parse codebases into AST knowledge graphs, enabling BFS dependency traversal for precise LLM context-stuffing. - Designed a Multi-Agent LangGraph Pipeline with a Supervisor intent router and parallel workers for automated code QA, 3-hop Blast Radius change-impact analysis, and PDF report generation. - Engineered Event-Driven Ingestion & Gemini Context Caching via Inngest; reduced LLM token costs by up to 80% using context caching while streaming real-time SSE progress to the Next.js frontend.
Key Engineering Milestones
Vectorless RAG & Graph-Native AST Engine
Architected a Vectorless RAG & Graph-Native AST Engine using FastAPI and Supabase to parse codebases into AST knowledge graphs, enabling BFS dependency traversal for precise LLM context-stuffing.
Multi-Agent LangGraph Pipeline
Designed a Multi-Agent LangGraph Pipeline with a Supervisor intent router and parallel workers for automated code QA, 3-hop Blast Radius change-impact analysis, and PDF report generation.
Event-Driven Ingestion & Context Caching
Engineered Event-Driven Ingestion & Gemini Context Caching via Inngest; reduced LLM token costs by up to 80% using context caching while streaming real-time SSE progress to the Next.js frontend.