Building and Scaling an AI Startup with Async Django
summary
Prototype RAG app uses async Django with Django Ninja and Pydantic, handles PDF upload → chunking → embeddings → vector store, and delivers streaming responses from LLMs via Django 4.2. MVP includes Swagger/OpenAPI, per-user auth and JWT, and localizable LLM/embedding hosting (vLLM suggested). Initial store is Chroma, then migrate Chroma → vector DB (or OpenSearch/Elastic for hybrid ranked retrieval). Scale path swaps blocking work to RabbitMQ/Celery workers, adds object storage, CI, tests and error reporting.
AI-written summary · about
chapters (11)
- 0:00Introduction and RAG concepts
- 2:07How RAG works
- 5:14Defining the MVP
- 8:49FastAPI versus Django
- 11:42Async Django implementation
- 13:35Frontend and testing
- 16:02Scaling the MVP
- 18:02Moving to production architecture
- 20:29Pipeline processing and workers
- 22:31Advanced retrieval and UI migration
- 24:12API automation and conclusion
related talks
45mDjango for AI: Deploying Machine Learning Models with Django with Will Vincent
47mBuilding Scalable Web Apps
29mSwiss Army Django: Small Footprint ETL
52mChallenges in developing a large Django site
49mBuilding AI Enabled Web Apps
26mProtoype for Production
30mBuilding modern Python web apps with PostgreSQL
38mQuerying Live Data With LLM App