Implementing an open-source RAG with OpenSearch
summary
An open-source RAG pipeline is implemented using Langchain for orchestration and OpenSearch as the vector store, with embeddings stored as 384-dimension vectors and retrieval done by vector nearest-neighbor. A local Llama.cpp-deployed model (from Hugging Face, fine-tuned ZIR) connects via an OpenSearch connector and agents to build prompts; the demo ingests documents with a pipeline, retrieves relevant docs, sends them to the LLM, and returns grounded answers — the local demo works end-to-end.
AI-written summary · about
chapters (9)
- 0:00Introduction to the team
- 1:18Limitations of LLMs
- 3:00Why LLMs fail as search tools
- 7:13Understanding RAG architecture
- 10:22Semantic search and vectors
- 15:13Open source RAG considerations
- 18:21Implementing RAG with LangChain
- 23:18OpenSearch connectors and agents
- 25:48Practical demonstration
related talks
25mSet Up an Open Source RAG in 30 Minutes
40mBuilding Intelligent Search Systems With Agentic RAG and OpenSearch
20mChallenges in Building RAG Systems (and Solutions Using OpenSearch)
22mOrchestrating Multimodal RAG With Agentic Workflows With OpenSearch
37mBuilding advanced RAGs with Elasticsearch OpenSearch
26mOpenSearch: Conversations with Context
01h 28mFrom RAG to Riches Building Enterprise Search Systems with LLMs
21mSmart Search with RAG: Elasticsearch Meets Language Models