UT Center for Identity · Research project
PrivacyCheck
A research project developed through the UT Center for Identity. It is archived and is not presented as an active Chrome Web Store product.
The problem
Privacy policies are long and difficult to compare. The project explored whether an LLM could turn policy text into a more useful, structured evaluation.
The approach
The extension collected policy text and sent it to AWS services for evaluation. Results were stored in DynamoDB so previously evaluated policies could be reused.
- Evaluate against explicit criteria
- Explain the score instead of returning one number
- Cache previous evaluations
The challenge
Policies vary widely in structure and detail. Producing consistent evaluations required clearer criteria than a general prompt.
What I learned
An LLM score is only useful when the criteria are visible. This project shaped how I think about structured evaluation and trustworthy AI products.