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: extracts the claims in a draft, searches arXiv / Semantic Scholar / PubMed / OpenAlex, scores source quality, and explains each recommendation. Usable from any MCP client.
- ResearchAssistantAgent — a chain of LLM agents that finds relevant papers, returns AMA-format citations, and ranks sources by a customizable rigor rubric.
- 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, rate-limited Python client for the OpenAlex scholarly-works API.
- scicite-claims — extract citation-worthy claim spans from scientific text with spaCy + linguistic patterns.
- Recommender-Systems — classic and modern recommender algorithms with a clean, typed, tested API.
Currently building toward agents that are right — and honest about when they aren't.