Intern: Matías Ignacio Pinto Chávez Project: OpenRegEval
OpenRegEval is a lightweight LLM evaluation agent for regulatory reasoning on FDA drug label content. It builds structured benchmarks, retrieves grounded evidence from openFDA and DailyMed, and scores model outputs for citation accuracy, factual grounding, and refusal behavior.
- MVP proposal and summary for a 2-week evaluation agent
- Full project plan for a summer-length OpenRegEval system
- Benchmarking and hallucination auditing workflows for FDA label reasoning
Python, PyTorch, Hugging Face Transformers, PEFT/LoRA, openFDA API, DailyMed API, pandas, scikit-learn, Streamlit