Indexes a whole codebase, its databases, and its datasets into one evidence-backed knowledge graph, then answers questions about how the system fits together with real file-and-line proof instead of LLM guesses.
- Full-stack platform (Node.js, Express, React) that statically indexes JavaScript, TypeScript, and Python repositories, SQL schemas, and CSV datasets into a cross-layer knowledge graph, tracing any value across seven layers from a frontend component down to the exact database column, with file and line evidence on every hop.
- Architected a multi-agent system (LangGraph, LangChain) where a supervisor routes each question to parallel specialist agents for repositories, architecture, and datasets, invoking typed tools validated by Zod schemas.
- Cut hallucinations with a deliberately non-LLM verification agent that re-checks every claim against the static index before composing an answer, labeling any hop it cannot confirm as unverified rather than inventing it.
- Deployed to production on Google Cloud Run (Docker, Artifact Registry, Secret Manager) with Firebase authentication and per-user isolated workspaces, engineered safe by construction: a read-only single-SELECT SQL gate, read-only database connections, and sandboxed upload handling.
7
layers traced end to end
3
parallel specialist agents
0
unverified hops passed as fact
Node.jsReactLangGraphLangChainGoogle Cloud RunFirebaseDocker