ResearcherX

Local-first AI research IDE with graph-based RAG and citation verification.


ResearcherX is a local-first AI research editor that helps researchers write against a bounded source library, using graph-based RAG to produce cited answers and flag unsupported claims, contradictions, or citation drift in drafts.

Highlights

  • Implemented the retrieval and verification pipeline behind the editor: turned PDFs, notes, and links into provenance-tracked source nodes, then checked draft paragraphs against relevant evidence with batched LLM calls.
  • Built a dual-routing pipeline with FastAPI and litellm to balance latency and API cost between local and frontier models.
  • Engineered asynchronous hybrid retrieval across Neo4j and LanceDB, with node-level provenance and Cypher-based garbage collection to preserve graph integrity and reduce hallucinations.

Tech Stack

Python (FastAPI), TypeScript (Next.js), Neo4j (Cypher), LanceDB, litellm, ProseMirror.