Case study — 03
Twy
Think freely. Twy organizes.
Sole Developer & Designerheytwy.com
A second brain you can talk to — typed, spoken or photographed input goes in unstructured and comes out clustered, with no folders and no setup.
- Problem
- Pen and paper is faster than any note app for capture, but a physical notebook turns into a mess within days — tasks, ideas and meeting notes mixed into one continuous stream, with finding something written last week feeling like archaeology.
- What I built
- Typed, spoken or photographed input goes in unstructured and comes out as clustered tasks and ideas — no folders, no tags, no setup. Hybrid semantic retrieval on pgvector with text-embedding-3-small handles the search half of that promise.
- Key decision
- Twy ships no speech code. It publishes tools and lets the Voice AI Platform call them over MCP — the payoff for building that platform as infrastructure rather than a feature.
- Also true
- Capture accepts text, voice or a photographed page via OCR — Twy reads handwriting as readily as it reads typed input, and works in any language.
Built with
- Next.js
- OpenAI
- OCR
- PostgreSQL