Real-Time AI with Open Source
summary
Builds a working RAG pipeline: pulls OpenAQ air‑quality REST data, ingests ~1,000 records at a time into a Milvus vector store, creates 384‑dim embeddings, and runs retrieval with LangChain calling a local LLM via Ollama and Python; demo loads ~60,000 of 100,000+ available records, uses Kafka and optional Flink for continuous ingest, shows IVFFlat indexing, multi‑vector multimodal search, metadata filtering and reranking, and returns coherent regional air‑quality reports — the demo mostly succeeds.
AI-written summary · about
chapters (15)
- 0:00Real-time AI intro
- 1:12Open air quality data
- 3:00Data ingestion challenges
- 6:14Unstructured sensor data
- 9:24Kafka and Flink streaming
- 12:15Fetching air quality data
- 14:06Schema for vector storage
- 16:36Querying vector embeddings
- 18:20RAG with LangChain Ollama
- 21:11Live air quality demo
- 35:05Time-series databases
- 36:50Domain data challenges
- 39:44Hybrid vector search
- 42:37Sharding and partitions
- 46:19Q&A
related talks
01h 30mIt's in the Air Tonight. Sensor Data in RAG
01h 35mTutorial: Build AI-Supercharged RAG Apps with a Vector Database
46mEvent Streaming and its Potential for the GenAI World
25mBuilding Advanced LLM Applications with Python and Milvus: A Scalable Vector DB Solution
46mAdvanced RAG with Azure AI Search - top questions answered
33mVector Databases in the Age of AI: Embracing RAG for Enhanced Search Capabilities
41mAdding Generative AI to Real Time Streaming Pipelines
16mOpenRAG: An open-source stack for RAG