Elastic, known for its search AI technology, has unveiled a new tool called Playground. This low-code interface allows developers to quickly create RAG (retrieval-augmented generation) applications with Elasticsearch. The tool's user-friendly design lets you perform A/B testing on various large language models (LLMs) and fine-tune retrieval mechanisms using your own data indexed into Elasticsearch.
“While prototyping conversational search, the ability to experiment with and rapidly iterate on key components of a RAG workflow is essential to get accurate and hallucination-free responses from LLMs,” said Matt Riley, Elastic's Global VP and General Manager of Search. He added that Playground brings together Elastic's comprehensive hybrid search capabilities and the innovation from various LLM providers, simplifying the process and speeding up time-to-market for customers.
Playground supports transformer models directly within Elasticsearch and is enhanced by the Elasticsearch Open Inference API. This API integrates models from a growing list of inference providers, including Cohere and Azure AI Studio.
Experiment and iterate with ease
Playground offers a straightforward interface for A/B testing different LLMs, such as those from OpenAI and Anthropic. It refines retrieval mechanisms to provide answers grounded in your indexed data. The tool leverages Elasticsearch's vector database and hybrid search capabilities, allowing you to use these features without added complexity.
Playground simplifies query construction by using query retrievers. With the latest version, you can perform hybrid searches using normalised scores. The tool understands the shape of your data and generates a unified query, enabling efficient searches. Soon, semantic re-ranking can be incorporated to improve recall quality further.
Once you've configured your semantic search, you can export the code and integrate it into your application using the Python Elasticsearch client or LangChain Python integration. Playground is now available on Elastic Cloud Serverless and in version 8.14 of Elastic Cloud.
