Fine-tuned scam message model, carefully evaluated

The brief

A phone app needs to flag scam messages on the device, in Indian languages. Fine-tune a small multilingual model, then evaluate it carefully: accuracy per language, false alarms on genuine messages, and how it handles new scam patterns. Good looks like: a clean multilingual test set, honest error analysis, and a comparison with a large hosted model.

What you will build

1. Build a labelled set in English and two Indian languages
2. Fine-tune a small multilingual model
3. Compare with prompting a large model
4. Evaluate per language and on new scam patterns
5. Report false alarms on genuine messages

Free, open data to build it with

• SMS Spam Collection ↗ UCI Machine Learning Repository

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.