Course project for Johns Hopkins University Graduate Level LLM Theory and Architecture course. Team members: Jieming Chu, Nainika Devireddy, and Amin Diab Course Instructors: Sam Barham and Ritwik Bose
The GitHub consists of our work for using LLM (GPT-4) for clinical trial recruitment. The proposal contains the original ideation. The MVP.ipynb is our initial attempts and exploration, trying different methods, prompt engineering, and querying techniques. The Synthea Exploration.ipynb is our initial attempts at pre-processing the patient file dataset and prompt engineering. Other files contain the output of the LLM, architecture graphs, and the final presentation slides.
We compared several approaches including feeding full patient EHR and clinicaltrials.gov eligibility data, summary data, and RAG-enabled approaches.