Search Toolkit

posts

2025–2026

Mistral AI introduces Search Toolkit, a composable, open-source framework for building production search pipelines for AI applications. It integrates ingestion, retrieval (BM25, dense embedding, hybrid), and evaluation (recall, precision, MRR, NDCG) into a single framework with a shared interface, aiming to reduce engineering time spent on plumbing and integrations. It supports various deployment environments (cloud, on-premises, edge) and is designed for enterprise search, RAG, and domain-specific applications. This post details using LLM-as-a-judge with the RAG Triad framework and Mistral's structured outputs to evaluate RAG systems, enhancing the evaluation capabilities within the Search Toolkit.

2026

Introducing Search Toolkit | Mistral AI

5/28/2026

This post introduces the Search Toolkit, a new framework for building AI search pipelines. It details the components of the toolkit: Ingestion (document parsing, chunking, embedding generation with adapter interfaces), Retrieval (BM25, dense embedding, hybrid configurations), and Evaluation (recall, precision, MRR, NDCG). It highlights the benefits of a unified framework for reducing integration overhead and improving search quality. The post also provides a starter app template with Docker and Vespa for quick setup and demonstrates its use with sample data. It mentions battle-tested use cases in financial services, manufacturing, public sector, and media & entertainment, including an example with CMA CGM and Voxtral for fake news detection.

2025

Evaluating RAG with LLM as a Judge | Mistral AI

4/9/2025

This post introduces a method for evaluating RAG systems using an LLM as a judge, specifically leveraging the RAG Triad (Context Relevance, Groundedness, Answer Relevance) and Mistral's structured outputs feature. It provides a Python code example using Pydantic models to define the evaluation schema for ContextRelevance, AnswerRelevance, and Groundedness, demonstrating how to structure LLM outputs for machine-readable evaluation. This enhances the evaluation component of the Search Toolkit by providing a concrete approach to assessing RAG pipeline performance.