Building RAG With OpenSearch ML Plugins: From PDFs To Voice-Enabled Search
summary
Shows how to deploy a sentence-transformer (multilingual MiniLM converted to torchscript) onto OpenSearch via the ML Commons plugin, extract text from PDFs with the ingest-attachment pipeline, chunk at 500 words with 50-word overlap, generate 384-d vectors inside the cluster, store them in KNN vector indices, and serve semantic search through a FastAPI layer. Whisper runs locally for STT and Mistral provides local LLM answers with citations. Demo indexes three chunks and returns top-K results with low latency.
AI-written summary · about
chapters (11)
- 0:00Building RAG with OpenSearch ML
- 2:30RAG system architecture
- 3:34OpenSearch ML Commons plugin
- 4:38Deploying embeddings model
- 7:47Processing PDFs
- 11:58Semantic search with KNN
- 13:22Generating answers with LLM
- 14:45Voice search with Whisper
- 15:27Live demo text & voice
- 18:13Key takeaways
- 20:59Resources & closing
related talks
38mBuild AI-Ready Search: Integrating Docling with OpenSearch for Advanced RA...
21mBuilding Privacy-First Search With Containerized OpenSearch and Local LLMs
30mBeyond Keywords: Building an AI-Powered Search That Reads Your Mind
26mRevolutionizing Search: Unveiling Conversational Search
26mOpenSearch: Conversations with Context
38mNatural Language Search-Lexical and Semantic
38mAI-Ready Data Foundation Using Amazon OpenSearch
51mQuery Understanding With Large Language Models: Techniques and...