The developer created a Go-based bot for Yandex.Telemost that automates meeting recording and transcription. The system uses WebRTC to join video calls as a participant, capturing audio and screen sharing tracks. It processes audio via ElevenLabs Scribe and generates summaries or mind maps using Gemini Flash 3.8 and Gemini Live.
The bot features an MCP server, allowing AI models like Claude and ChatGPT to access meeting transcripts and summaries. It also integrates with Telegram and supports authentication via corporate email using JWT tokens. The backend is built with Go, utilizing libraries such as chi, pgx, and pion/webrtc.
A key feature allows users to join video meetings via a standard phone call. By dialing a number and entering a code, the user connects to the meeting as a live participant. The system uses sipgo for signaling and ffmpeg to handle real-time audio transcoding between 8kHz PCMU and 48kHz Opus formats.
The project evolved from a simple Python script for audio transcription into a complex tool. The developer addressed technical challenges such as session ID persistence in WebRTC, handling VP8 screen capture artifacts, and implementing debounce logic for DTMF signals to prevent duplicate input during phone-based meeting access.
