Research Infrastructure

Search, memory, and agent tooling I built to accelerate my own research — it powers the citation-graph demo.

research-kb

in progress

A hybrid-search knowledge base I built to accelerate my own research: BM25 full-text, vector embeddings, and a citation and concept graph over 478 sources (236K chunks), focused on causal inference. It's exposed as MCP tools so an assistant can search and traverse it directly, and it's the data source behind the /lab/research-graph demo on this site. Infrastructure for my own learning, not a product.

Stack: Python · BM25 · vector search · knowledge graph · MCP

What's next

Expanding corpus coverage beyond causal inference (time-series, econometrics) and densifying the citation graph.

research-agent

in progress

A multi-agent research pipeline built on research-kb: it decomposes a question, retrieves evidence, synthesizes a structured answer, and validates citations against the sources. It's the querying layer of the same enabling system — how I turn the knowledge base into vetted answers.

Stack: Python · LangGraph · MCP · research-kb

What's next

Tightening synthesis quality and wiring outputs back into research-kb for cross-topic reuse.

research_toolkit

in progress

A collection of Claude Code skills exposing the research-kb and research-agent stack as repeatable workflows: literature review, causal-assumption audits, citation-network analysis, and cross-domain concept discovery. The point is to validate research claims systematically before anything is published.

Stack: Python · Claude Code · MCP · research-kb · research-agent

What's next

A methodology page on how that validation works — the local research-* dogfooding sandboxes that vet claims before they ship — is coming under this cluster.