Mini-project
Pharma & Life Sciences
Adverse event extractor for pharmacovigilance
Pull drug, reaction and severity from safety reports into validated data, and rank the most reported.
The brief
Pharmacovigilance teams read thousands of safety reports to spot side effects early. Build a pipeline that extracts drug and reaction pairs and severity from report text into validated JSON, and ranks the most reported pairs. Good looks like: extraction accuracy on a labelled corpus, a comparison of an NER model with an LLM, and a ranked table with counts.
What you will build
- Explore the adverse-event corpus and report data
- Extract drug and reaction pairs with an NER model
- Extract the same with an LLM and structured outputs
- Compare accuracy on labelled sentences
- Rank the most reported drug and reaction pairs
Free, open data to build it with
- ADE corpus v2 ↗ Hugging Face
- openFDA adverse event reports (FAERS) ↗ U.S. Food and Drug Administration
Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.