Harnessing DeepSeek-R1 and OpenSearch for RAG-Powered Intelligent Info...
summary
Built a RAG pipeline that connects DeepSeek-R1 on SageMaker (generation) with a lightweight Hugging Face sentence-transformer for embeddings and OpenSearch vector DB (v2.17) via ML connectors, ingesting documents as KNN vectors and serving a chatbot that reasons over data. Demo returns concise, inferred answers (example: Bangalore overtakes Mumbai ~2037, Hyderabad ~2047), reports higher Q&A accuracy than alternatives, and uses KNN k=5 size=2 for retrieval; scaling relies on sharding, replication, GPU instances, caching, and distributed queries.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:01Enterprise search challenges
- 3:56Metadata-driven search limits
- 4:38What is RAG workflow
- 7:31RAG demo overview
- 7:52Vector databases and LLMs
- 10:20RAG ingestion and generation
- 11:49Embedding vs generation models
- 12:11DeepSeek for RAG generation
- 14:17OpenSearch as vector DB
- 16:02OpenSearch vector use cases
- 17:07ML connectors in OpenSearch
- 21:20Demo solution architecture
- 22:31Live RAG chatbot demo
- 31:21Scaling RAG strategies
- 32:47Key takeaways
- 34:42Q&A
related talks
38mAI-Ready Data Foundation Using Amazon OpenSearch
34mImplementing an open-source RAG with OpenSearch
40mBuilding Intelligent Search Systems With Agentic RAG and OpenSearch
21mSmart Search with RAG: Elasticsearch Meets Language Models
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!
45mMaking semantic search & RAG real: How to build a prod-ready app
25mSet Up an Open Source RAG in 30 Minutes
31mOrchestrating Multimodal RAG With Agentic Workflows