Brag Your RAG with the MLOPS Swag
Talkat Cloud Native + Kubernetes AI Day NA 2024 · on Tuesday, 2024-11-12· at 14:40· Salt Palace | Level 1 | Grand Ballroom A·
summary
Live demo builds three RAG paths: LangChain + Pinecone, Canopy + Pinecone, and a pure LangChain variant without a vector DB; all deploy on Kubernetes and the Canopy+Pinecone flow successfully answers a query about the 2024 Olympics and returns its source, proving grounding fixes hallucinations. Embedding pipeline runs with Ray on GKE, stores vectors in PostgreSQL/PGVector on Cloud SQL, streams data with GCS Fuse to avoid idle GPUs, and targets multi‑GPU A3 nodes plus secondary boot disks and service extensions for efficient serving.
AI-written summary · about
chapters (11)
- 0:00Introduction and agenda
- 1:14LLM customization spectrum
- 3:49Prompt engineering and RAG
- 6:58Open source RAG implementation
- 8:36RAG demo and comparison
- 12:18Technical code walkthrough
- 13:23MLOps pipelines on GKE
- 16:00Deep dive into embeddings
- 17:11Optimizing storage with FUSE
- 18:07Advanced training patterns
- 24:14Q&A and session closing
related talks
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
31mBuild an Agentic RAG system using Langchain, Ollama and Milvus
02h 47mFrom naive to advanced RAG: the complete guide
36mProduction Ready RAG
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
25mComposable RAG Pipelines With OPEA Highlighting GraphRAG
27mStreamlining RAG Applications with Pinecone's OSS Framework: Canopy
25mBuilding Advanced LLM Applications with Python and Milvus: A Scalable Vector DB Solution