Relevance workbench
In this workbench, you can compare our Elastic Learned Sparse Encoder model (with or without RRF) and traditional textual search using BM25.
Start comparing different hybrid search techniques using TMDB's movies dataset as sample data. Or fork the code and ingest your own data to try it on your own!
Try these queries to get started:
- "The matrix"
- "Movies in Space"
- "Superhero animated movies"
Notice how some queries work great for both search techniques. For example, 'The Matrix' performs well with both models. However, for queries like "Superhero animated movies", the Elastic Learned Sparse Encoder model outperforms BM25. This can be attributed to the semantic search capabilities of the model.
Explore similar demos
Security
Elastic Security: Agentic Security Operations Platform
Step through Elastic’s agentic SOC live. Watch autonomous agents correlate and investigate a real alert end-to-end, then stage a response through Elastic Workflows—with deterministic playbooks and agentic reasoning working side by side. You stay in control at every gate, approving agent actions and seeing exactly how judgment and verification remain with the analyst while the busywork runs itself.
Search
Elasticsearch Vector Database - Getting Started
Use Elastic Serverless to create an Elasticsearch Vector Database project. No infrastructure to manage. We add data, generate embeddings, and get started fast with code samples you can run in the Elastic Cloud console.
Search
Elasticsearch Vector Database - Python & PDF in S3, Claude Code
Walk through the steps required to create a Vector Database and how to use Python code to index a PDF document from an Amazon S3 bucket into a Vector Database index. Then we'll create an Elastic workflow and MCP tool that can be queried by Claude Code to give it custom context.