AI Codebase Intelligence Platform

Problem

Large multi-language codebases are hard to search by meaning and reason about — which code is relevant, where a change ripples, and what's risky to touch.

What I built

• Architected a hybrid RAG system (Qdrant vector search + a code graph) for semantic search across 80+ languages.
• Built multi-language AST parsing, Git-history risk scoring, and tenant-isolated embeddings.
• Shipped LLM-guarded retrieval with blast-radius analysis and a D3.js dependency-graph view.

Impact

• Semantic code search across 80+ languages from a single query.
• Surfaces change blast-radius and risk before edits land.

Stack