Skip to content

Case study 12

MustBreak

MustBreak, built at MUST Company in Seoul, is a real-time AI translator for meetings and chat. People speak in their own language and everyone in the room reads the conversation in theirs, with automatic language detection, speaker identification and noise cancelling. Meetings become shareable minutes with speaker separation, AI summaries and chat, and searchable history, so late joiners can catch up. In Slack, the @MustBreak bot translates conversations in place.

Visit live site

At a glance

Status: Completed

Role
Full-stack engineer
Type
Enterprise
Timeline
May 2024 – Jun 2026
Domain
AI, SaaS
Focus
Real-time Translation, Speech Transcription, Multilingual Collaboration, Slack Integration
Stack
  • Python
  • FastAPI
  • DynamoDB
  • Next.js
  • TypeScript
Architecture: Real-time translationMeeting app to Realtime API (audio stream); Realtime API to AI models (transcribe · translate); Realtime API to DynamoDB (transcript); Slack to Realtime API (events); DynamoDB to Meeting notes (summarize)audio streamtranscribe · translatetranscripteventssummarizeUIMeeting appNext.js · mobile webSVCRealtime APIFastAPIEXTAI modelsspeech · translationDBDynamoDBtranscripts · historyUISlack@MustBreak botJOBMeeting notessummaries · chat
Fig. 12 MustBreak architecture: Real-time translationSolid: request · dashed: async · dashed box: external

What MustBreak does

  • Real-time transcription and translation across 60+ languages
  • Group conversations with speaker identification
  • Automatic language and voice recognition
  • Noise cancelling that isolates the speaker's voice
  • Pronunciation support for non-native English speakers
  • Meeting notes with speaker separation, AI summaries and AI chat
  • Searchable conversation history, including catch-up for late joiners
  • Shareable, confidential meeting minutes
  • @MustBreak Slack bot for translation inside Slack
  • Mobile web and mobile app access

The problem

Distributed teams lose meetings to language. Non-native speakers follow half the conversation, notes miss what was said, and the same friction carries over into chat.

Approach

Audio streams from each participant are transcribed and translated in real time, tagged by speaker and fanned out to everyone in their chosen language. The same transcript then drives meeting notes, AI summaries and searchable history, and a Slack bot brings translation into the team's chat.

Key decisions

  1. D1

    Latency first

    Translation is only useful while the conversation is still happening, so the real-time path is kept short and low-latency from microphone to every listener.

  2. D2

    One transcript, many uses

    Live captions, minutes, summaries and search all read from the same speaker-tagged transcript.

  3. D3

    Meet teams where they talk

    The Slack bot brings the same translation into chat, so it isn't limited to meetings.

Highlights

  • Developed MustBreak, an AI-powered real-time meeting platform for transcription, translation and multilingual collaboration across distributed teams, with AI-generated summaries and searchable conversation history.
  • Integrated Slack workflows and low-latency real-time communication infrastructure to support enterprise-grade multilingual collaboration.

My contributions

  • Worked across the stack as a full-stack engineer at MUST Company
  • Built low-latency real-time transcription and translation infrastructure
  • Built the @MustBreak Slack bot and the Slack workflows around it
  • Built meeting features: speaker separation, AI summaries and chat, and searchable conversation history
  • Built responsive frontend features in Next.js and TypeScript

Build something similar

Planning a product like MustBreak? Tell me what you're building and when you need it. I'll reply with questions or a proposal.

Request a quoteHire me on Upwork