How attendee-labs/attendee works
Project type and purpose
• Attendee is a full-stack Django monolith that exposes a REST API for launching and managing “meeting bots” on Zoom, Google Meet, and Teams. It includes a web UI (Django templates + static files), an API layer (Django REST Framework), background workers (Celery), and adapters to external services (video conferences, transcription providers, payment processing).
File structure
• Top-level:
• attendee/ – Django project settings, URL routing, WSGI/ASGI entrypoints
• accounts/ – standard Django Auth + signup/login/email confirmation
• bots/ – core “bots” app
– adapters (Zoom, Google Meet, Teams, web hooks)
– controller (media pipelines, audio/video managers)
– bot pod creation logic (Kubernetes integration)
– tasks (Celery jobs for launching, running, transcription, webhooks)
– serializers, API views, templates for the project UI
• Docker & deployment files (Dockerfile, dev.docker-compose.yaml, heroku.yml, Procfile)
• Configuration (pyproject.toml, requirements.txt, Makefile, .github/workflows/ci.yml)
• Documentation (docs/, README.md)
README for additional insights
• Single Docker image, requires only Postgres and Redis
• Uses Zoom OAuth, Chrome-based Google Meet, Teams SDK
• Transcription providers like Deepgram, AssemblyAI, etc.
• Celery for async tasks, Stripe for billing, AWS for storage (via env vars)
Main components
a. Client Applications
– Developers calling the REST API (/api/v1/bots, /bots/<id>/transcript)
– Web UI (Django templates + JS) for managing API keys and credentials
b. Django Web App
– Gunicorn/Uvicorn serving both UI and JSON API
– Authentication layer (Token auth for API)
– bots app controllers & views
c. Database & Cache
– PostgreSQL (state, users, credentials, transcripts, billing)
– Redis (Celery broker + cache/webhook queue)
d. Background Workers
– Celery workers running tasks:
• Launching bot pods via Kubernetes API
• Managing streaming pipelines (gstreamer, Chrome drivers)
• Uploading media artifacts (to S3 or similar)
• Dispatching transcription jobs
• Sending webhooks on state changes
e. Bot Pod Infrastructure
– Kubernetes cluster (GKE) or Docker-Compose local pods
– Pod template containing the bot controller code + media pipeline
f. External Services
– Video Conference Providers (Zoom API, Google Meet via Chrome, Teams SDK)
– Transcription Providers (Deepgram, AssemblyAI, OpenAI Whisper, Google TTS)
– Payment/Billing (Stripe)
– Optional S3-compatible storage for recordings
g. Webhooks & Callbacks
– Outbound HTTP webhooks configured per project on state/transcription events
Relationships and interactions
1. Client → Django API (HTTP)
2. Django enqueues Celery task to launch a bot
3. Celery worker → Kubernetes API → new Bot Pod
4. Bot Pod connects to video service (Zoom/Meet/Teams)
5. Bot Pod streams audio/video through gstreamer pipeline
6. Bot Pod uploads media chunks to storage or relays them to transcription task queue
7. Celery worker picks up transcription tasks → calls external transcription API → stores transcript in DB
8. Celery worker sends webhook or updates DB state
9. Client polls or receives webhook and fetches transcript
Architectural patterns and design principles
• Monolithic Django codebase split into “apps” for modularity (accounts vs. bots)
• Asynchronous task queue (Celery + Redis) for long-running operations
• Adapter pattern for supporting multiple video and transcription providers
• Kubernetes as the execution environment for ephemeral, isolated bots
• Single Docker image for consistent deployment, with environment-specific settings
Detail
• Environment variables/config (e.g., Zoom client ID, Deepgram key)