← Work
Experiment2026 · Solo Builder· Started(First Commit date)
Lecturenote AI
Pairs lecture slides with a recording to draft per-slide study notes - Whisper STT + Claude.
- React
- FastAPI
- Whisper
- Claude
Build
- What I did
- Renders each slide as an image and matches it to the voice transcript at slide granularity
- Claude drafts lined notes beneath each slide
- QA considerations
- When aligning the STT transcript to slide boundaries, an offset drift doesn't attach notes to the wrong slide
- Matching stays robust across audio edge cases - Whisper misrecognition, long silences, overlapping speakers
- Claude's note draft doesn't invent content absent from the transcript (guaranteed a draft, not a fact)
- For a slide passed over with no speech (an empty transcript segment), it leaves the note blank rather than forcing one
- How to measure per-slide note accuracy and prevent regressions - without a metric this becomes "keep building without deciding," the open gate of this project
- Tech stack
- React
- TypeScript
- Vite
- FastAPI
- Python
- Whisper
- Claude