← All projects
Feb 2026
TwinMind Copilot
A live meeting assistant that transcribes speech in real time and suggests what to ask next.
TwinMind Copilot listens to microphone audio, transcribes it in real time, generates actionable suggestions (questions to ask, fact-checks, talking points), and offers a context-aware chat to dive deeper into anything discussed.
Highlights
- Real-time audio: 30-second microphone chunks transcribed with Groq's
whisper-large-v3, with suggestions arriving in under 5s on average. - Contextual decision engine: Groq's
gpt-oss-120bclassifies the transcript into 5 insight types (Questions, Talking Points, Answers, Fact-Checks, Clarifications) and generates 3 suggestions per cycle. - Context-aware chat: click any suggestion for a markdown deep dive; a configurable 15K–40K character transcript window and 30-message history cap keep token payloads small.
- Split deployment: React/Vite on Vercel, Dockerized Django REST API on Fly.io, locked down with a CORS allowlist of 2 origins. Shipped in 12 PRs over 4 days.