Build an Agentic RAG system using Langchain, Ollama and Milvus
summary
Live demo builds an agentic RAG that combines LangChain 4J, Ollama Llama 3.2 (running locally) and Milvus vector DB, plus Tavi for web search. Pipeline embeds docs with MiniLM, stores chunks in Milvus, routes queries (vector vs web), grades retrieved docs, adds self-reflection and memory, and reissues searches when answers hallucinate. Claims scale to billions (customers at ~100B vectors) and shows the end-to-end system functioning on a laptop Docker setup.
AI-written summary · about
chapters (8)
- 0:00Introduction to Milvus
- 3:46Vector database mechanics
- 7:31Understanding vector embeddings
- 11:33Retrieval augmented generation
- 13:44Limitations of naive RAG
- 16:47Moving to agentic RAG
- 19:19Live demo and implementation
- 27:47Advanced agent workflows
related talks
25mBuilding Advanced LLM Applications with Python and Milvus: A Scalable Vector DB Solution
48mGoing Beyond Prompt Engineering
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
34mImplementing an open-source RAG with OpenSearch
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
29mPrivacy-Friendly Applications with Ollama, Vector Functions, and LangChainJS
29mUsing LangChain and LlamaIndex to customize LLMs...
25mSet Up an Open Source RAG in 30 Minutes