NTULearn toolkit
Turns my university lecture recordings this semester into study notes.
- role
- Designed and built it for my own studies.
- stack
- Python · Playwright · FFmpeg · SQLite · Claude Code CLI
This semester I'm taking an overloaded course load to graduate early: 5 senior-level maths modules and 1 scientific communication module. The toolkit keeps up with my weekly lectures, and together with my home server's morning brief and weekly school review, it keeps track of assignments, important dates and upcoming exams.
A 2-hour lecture becomes about 100 slide images, which a vision model reads into notes.
The model was chosen by an A/B test that caught the cheaper one inventing content.
Frames sampled from a lecture recording are fingerprinted, and near-duplicates are dropped, leaving about one image per slide. A vision model reads those into notes, while the transcript feeds an evening digest.
The model was picked by running both candidates over the same 99 frames and checking every disagreement against the slides themselves.
build notes
- The cheaper model made things up · Choosing a vision model by testing both on the same 99 slides.
Built with Claude Code multi-agent workflows: I designed the architecture, owned the decisions, and verified the output; agents wrote most of the code.