Clinical Application¶
Seven mini-projects, ascending in complexity, building tools grounded in a real clinical workflow: patient-facing chat, note summarization, triage, drug safety, radiology review, and trial matching. Builds on the general-purpose skills from Foundations. A frontend built by prompting an AI coding agent (Claude Code, Codex, or VS Code agent mode), talking to a backend built on open-source models and datasets. Early projects are pure prompting; later ones introduce fine-tuning a small open model.
| # | Project | Backend | Training? |
|---|---|---|---|
| 1 | Appointment Prep Chatbot | Open LLM + RAG over an FAQ doc (Chroma) | No |
| 2 | Patient Note Summarizer | Open summarization model / LLM prompting | No |
| 3 | Symptom-to-Specialist Router | Fine-tuned DistilBERT classifier | Yes |
| 4 | Drug Interaction Checker | RxNorm dataset lookup + LLM explainer | No |
| 5 | Radiology Report Flagger | Fine-tuned NER model | Yes |
| 6 | Clinical Trial Matcher | ClinicalTrials.gov API + sentence-transformer ranking | No |
| 7 | Capstone: Personal Health Assistant | LoRA fine-tune + RAG over uploaded docs | Yes |
Every project's backend runs on open-source models and datasets, and everything is inspectable locally. New to the tooling in general? Start with Software Engineering Practices for Data Science.