AI / agent engineer. I build LLM agents and tooling for work where being confidently wrong is expensive: clinical research and quantitative finance. So my projects lean hard on reproducibility, evaluation, and knowing the limits of an answer.
Background: peer-reviewed clinical-data analysis, live options-trading systems, and production ML for healthcare, biotech, and finance clients. Now focused on agents you can actually trust to deploy.
New York, NY / San Diego, CA (bicoastal) · databurton.com
- ResearchAssistantMCP: an MCP server for citation-finding research. It extracts the claims in a draft, searches arXiv, Semantic Scholar, PubMed and OpenAlex, scores source quality, and explains each recommendation. Usable from any MCP client.
- statskeptic: a data-analysis
agent that red-teams its own conclusions and reports what the data can't support.
pip install statskeptic
- csvtriage: load messy CSVs (bad encodings, ragged rows, junk), recover what's parseable, and report every decision instead of failing silently.
- oalex: a typed async client for the
OpenAlex scholarly-works API, with API-key auth, disk caching, and credit-aware
retries.
pip install oalex - scicite-claims: extract citation-worthy claim spans from scientific text with spaCy and linguistic patterns.
- lead-gen-pipeline: scrape Chamber of Commerce member directories into structured business records, with a crawler that identifies itself and honors robots.txt.
- Recommender-Systems: classic and modern recommender algorithms behind one typed, tested API, benchmarked on MovieLens, with Rust kernels.
- cinestyle: film-inspired
matplotlib themes with a colorblind audit that repairs the palettes it flags.
pip install cinestyle - echo-translate: hear English spoken in another language in your own cloned voice, offline, for pronunciation practice.
Currently building toward agents that are right, and honest about when they aren't.