Knowl
Knowl is a Cursor-style workspace where students work with full AI assistance while every prompt, draft and rubric score gets logged for evaluator review. I built it in 24 hours at the Michigan AI Hackathon with a CS undergrad and an MBA, and it was fully functional end to end by the deadline.
- Role
- Built the front end and all AI features
- Timeline
- February 2026, 24 hours
- Stack
- Next.js, TypeScript, Prisma, Claude API, Monaco, Tiptap
The problem
If students are allowed to use AI, grading the output tells you almost nothing. The thing worth evaluating is the process: what they asked, what they changed, and what they accepted.
What I built
I built the workspace layouts, the Monaco and Tiptap editor integrations, an AI chat panel with cursor-style edits and suggestions, and the evaluator review flow, which shows prompt traces, draft diffs and AI-suggested scores with the reasoning visible.
What I'd do differently
It's a hackathon project and I treat it as one. What I took from it is that logging the full trace made the AI legible to a third party, and that turned out to be the feature. I saw the same thing later at Insora, where the audit trail ended up carrying the product.