Research Engineering Intern, Open-Source Healthcare AI
Remote, anywhere6 months~20 hours/week
Help build an open benchmark and toolkit that measures whether AI can correctly apply real Medicare coverage policy, and the open data and developer tools that go with it.
About the project
Medicare decides what it pays for using hundreds of long, frequently updated policy documents. Doctors, billing teams, insurers and a growing number of AI products all have to interpret them, and a wrong call means a denied claim or delayed care. AI is already being used for these decisions, but there is no public, rigorous way to measure whether it gets them right.
This project builds that measure: an open benchmark grounded in the real CMS coverage policies, plus a machine-readable criteria dataset and open-source tools developers can build on. Everything ships publicly on Hugging Face and GitHub, with a paper on arXiv.
What you'll do
- Turn CMS National Coverage Determinations into structured, source-linked criteria checklists
- Build and validate synthetic clinical cases with verifiable coverage decisions, working with clinical reviewers
- Write the evaluation harness and run open models (via vLLM) and frontier APIs against the benchmark
- Build an open-source MCP server and API so AI agents can search and apply coverage policy
- Publish datasets and models on Hugging Face with clear dataset cards and documentation
- Contribute to the accompanying research paper and a public leaderboard
What you'll get
- Weekly 1:1 mentorship on AI infrastructure, evaluation and research
- Co-authorship on the paper and public credit on every release
- A visible open-source portfolio with real users
- A recommendation letter based on your work
What you'll need
- Strong Python, and comfort working with LLM APIs or open models
- Careful reading of dense technical or legal text, and attention to detail
- Clear written English for documentation, dataset cards and the paper
- Ability to work independently across time zones and to own tasks end to end
Nice to have
- Background in healthcare, medical billing and coding, nursing, or health policy
- Experience with vLLM, Hugging Face datasets, or building evaluation benchmarks
- Experience with MCP, agent tooling, or open-source contributions
- Prior research writing or a published paper
How to apply
Email manoj.dobbali@gmail.com with the subject “Research Engineering Intern application” and include:
- Your resume or LinkedIn, and your GitHub
- Two or three sentences on why this project interests you
- A short exercise: read CMS NCD 210.14 (Lung Cancer Screening with LDCT) and describe one patient scenario where deciding coverage would be tricky, and why
This is an independent open-source project. It is not affiliated with or endorsed by CMS or any employer.