Symptom coding model, carefully evaluated

The brief

Triage nurses turn patients' descriptions into symptom codes. Fine-tune a small model to suggest the codes from a patient's own words, then evaluate it carefully: accuracy per code, errors that would matter, and whether it does worse for some kinds of description. Good looks like: a clean test set, error analysis with clinical input, a bias check, and a human-review step.

What you will build

1. Prepare descriptions with symptom or condition labels
2. Fine-tune a small model with LoRA
3. Compare with prompting a large model
4. Analyse errors and check performance across groups
5. Design the human-review step

Free, open data to build it with

• Symptom2Disease ↗ Kaggle

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.