Integrated Multi-Modal Search: Generate and Search Image Vectors Directly in Elasticsearch

## Value proposition

By integrating multimodal models into the Inference API, we are changing what "search" means for a developer using Elastic. For the first time, developers can generate multi-modal vectors within Elastic for applications that perform sophisticated multi-modal search, like searching images with text, by using the same single, seamless and intuitive Elasticsearch Inference API syntax they already know.

This integration allows developers to build, deploy, and scale applications that can see, read, and understand complex, real-world information in Elastic, rather than having to bring their image and multi-modal vectors from external platforms.

## Expected outcome

MVP to generate multi-modal vectors (text and image) in Elastic to enable multi-modal search.

**Example use cases:**

- E-commerce: An online fashion retailer allows customers to search through clothing and accessories using natural language queries. It recommends items based on visual similarity (image-to-image) alongside similarity based on natural language descriptions and positive reviews by customers with similar profiles.
- RAG for customer and technical support: A manufacturer has thousands of technical documents and manuals involving: text and technical diagrams with functional and technical descriptions and operational instructions. Customers or technicians can ask questions or for troubleshooting steps and the response may contain instructions with diagrams.
- An AI search engine for researchers has thousands of research publications. Researchers describe a topic in natural language and the tool returns an overview, retrieving the most relevant articles and understanding charts, text and tables.