Project 1: Medical Term Definer¶
Mentors: To be announced
Problem: Clinical notes are full of jargon and abbreviations a patient can't parse on their own, and there's no lightweight way to surface plain-language definitions without rewriting the note itself.
Context: Runs on mtsamples medical transcription reports as test notes, using a local open-source LLM via Ollama (e.g. Llama 3 8B), prompting only, no fine-tuning.
Goals: Given a clinical note, flag jargon/abbreviations a patient wouldn't know and generate a one-line plain-language definition for each, without altering the rest of the note.
Deliverables: A single-page app: a textarea in, a rendered note out with <mark> spans over jargon terms (tooltip or footnote holds the definition), with definitions coming from one LLM call per note, constrained by a system prompt to a term→definition JSON list that the frontend merges back into the note text.
Showcase: TBD
References:
- MedJEx: A Medical Jargon Extraction Model with Wiki's Hyperlink Span and Contextualized Masked Language Model Score (EMNLP 2022): the closest academic match, a model trained specifically to flag jargon terms in EHR note sentences, the same task this project prompts an LLM to do.
- README: Bridging Medical Jargon and Lay Understanding for Patient Education through Data-Centric NLP: pairs jargon detection with retrieved lay definitions, the same two-step shape (flag, then define) as this project's deliverable.
- OpenNotes: the research movement behind patients reading their own clinical notes; this project is a lightweight version of the comprehension problem OpenNotes documents at scale.