An AI assistant that helps support teams answer policy questions
- What was built
- A retrieval-based assistant over a policy library, with a hand-off to a human when confidence is low.
- How it was tested
- An evaluation set of real questions, accuracy and hand-off rates, and a red-team pass for unsafe answers.
- Learner's role
- Product design, retrieval set-up and evaluation; worked in a pod of four.
- Gate 01
Problem and evidence
- Gate 02
AI suitability and concept
- Gate 03
Process, UX and architecture
- Gate 04
Prototype or proof of concept
- Gate 05
Evaluation and user evidence
- Gate 06
Red team, security and responsible AI
- Gate 07
Governance and economics
- Gate 08
Final board defence, live