Scalable Open AI GPT Applications in the Cloud
summary
Live-coding builds a React+Node lodging recommender that starts by calling GPT-4 chat completions (10–15s latency) then switches to text-embedding-ada-002 embeddings stored in Postgres with pgvector, cutting response time to under a second. 7,000+ Airbnb descriptions produce 1,536‑dim vectors (embedding generation ~15 minutes), indexed with an HNSW index for ANN search; when dataset outgrows a single node the app runs on distributed Postgres (YugabyteDB) for horizontal scale and sharded reads/writes.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:52What is OpenAI
- 8:09First app version
- 12:39Embeddings and PostgreSQL
- 22:36Optimizing vector indexes
- 34:15Scaling with distributed DBs
- 42:01Summary and deployment
- 43:50Q&A
related talks
45mUsing Postgres + OpenAI to power your AI Recommendation...
31mText and Code Embeddings
47mExtendable by Design: Building Generative AI Apps with Postgres and Vector Storage and Azure AI
15mHarnessing the Power of Generative AI for Intelligent Applications
15mThis or That with Takara AI
51mEvent Driven AI: Supercharging ChatGPT with RAG & LangStream
28mWhat the RAG? Demystifying Retrieval Augemented Generation
37mBuilding Generative AI Applications