How leo-dev-2004/team-project-context-translator-ss25 works
Project type and purpose
• This is a full-stack, real-time virtual meeting assistant built as a browser extension/plugin for Google Meet, backed by two specialized microservices (STT and LLM).
• Its purpose is to capture live audio, convert speech to text, use LLMs to generate contextual explanations, summaries, and suggestions, then render those insights back in the meeting UI.
File structure
• Top-level:
• .github/workflows/summary.yml – CI workflow definition (build/tests)
• .gitignore – ignores local/dev files
• README.md – full project spec and architecture overview
• No src directories present yet, but the README defines three independent modules/groups (App, STT, LLM) that will map to separate code repos or sub-folders in src/.
Key insights from the README
• Three core components:
1. App Group (Frontend): Google Meet extension -> capture/display
2. STT Group: real-time speech-to-text microservice
3. LLM Group: inference microservice for explanations, summaries, suggestions
• Data flow:
Audio → STT → sentence splitting → LLM → UI
• Technologies: Chrome/Firefox WebExtension APIs, Hugging Face STT models, Qwen-family LLMs, WebSocket/HTTP for comms, optional GPU workstation or cloud GPU.
Detail
• Every step: audio capture, streaming, segmentation, inference, display.
• Both cloud/GPU compute and on-device elements.
• Don’t omit the CI/CD workflow even if it’s peripheral.