Our work / StudyLamp
StudyLamp: an AI study partner that follows your unit’s rules
An AI study partner for Australian university and TAFE students. It answers from their own course files, and is in pre-launch testing.
The problem
Each unit sets its own rules on AI, and students are left to work out what’s allowed. General chatbots don’t know those rules, can’t show where an answer came from, and leave no record.
What we used to build it
- Interface
- React, Vite and Tailwind CSS, in TypeScript, installable on a phone
- Server
- Hono on Node.js with Better Auth, hosted on Fly.io in Sydney
- Data
- Postgres with pgvector on Neon in Sydney, with row-level security so each student’s data stays theirs
- AI
- OpenAI and Anthropic models through the Vercel AI SDK, with routing between them
- Files
- Course files stored in Cloudflare R2, with Cloudflare DNS
- Testing
- Evaluation suites that run the real model against fixed cases
The choices that make it responsible
Rules are checked in code
If a unit forbids AI-written text, StudyLamp won’t draft or rewrite the work. That is decided by tested rules before an answer reaches the student, not left to the AI.
Answers show their source
Answers cite the page they came from, and say so when the files don’t cover a question.
Limits are visible
Every answer carries a label: Clear, Flag, Check or Blocked. Students can challenge an answer, and the tutor checks again.
A record students can use
Questions and answers are kept with the rules in force at the time, and become an AI-use declaration a lecturer can check.
What we’re still improving
Evaluations already cover retrieval, rule-following and quiz quality. Fairness testing is still to come.
Want something like this for your organisation?
Tell us what you’re trying to do. If AI isn’t the answer, we’ll say so.