Sygmatic
Authentic language learning from real YouTube videos.
The problem
Textbook audio is sanitized and unnatural. Learners want to understand how the language is actually spoken — at real speed, with real prosody — but raw native video has no scaffolding to learn from.
The approach
Sygmatic builds lessons on top of real native-speaker YouTube clips: every word is transcribed and click-to-seek, every clip is rendered as an interactive waveform, and each lesson ends in a quiz generated from the speech's own part-of-speech and tense metadata. The clever bit on performance is that waveform peaks are precomputed offline and committed to git, so Wavesurfer v7 renders instantly without re-analyzing audio in the browser. Originally a paid SaaS at sygmaticlanguage.com, it was modernized in 2026 on Next.js 16, Drizzle on Aurora Serverless v2, Auth.js v5, and SST 4.
Highlights
- 100+ authentic video lessons across English, Spanish, German, French
- Word-level transcript with click-to-seek and per-word definitions
- Waveform peaks pre-generated from YouTube audio and committed to git
- Quizzes auto-generated from each lesson's part-of-speech and tense metadata
Tech stack
Want to talk through how this was built?
I build and ship production apps end-to-end. Happy to dig into the architecture, the tradeoffs, or your next project.