Foundry IQ: Improve recall by up to 54% with knowledge bases

2026-06-02 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, long

Summary

Foundry IQ knowledge bases significantly enhance agentic retrieval performance, improving evidence recall by up to 54% and reducing token costs by 34% compared to standalone retrieval tools. Evaluated on the BrowseComp-Plus benchmark, replacing single-shot RAG with a knowledge base boosts evidence recall by up to 46%. These improvements stem from a dynamic agentic retrieval loop, an enhanced semantic ranker, better answer synthesis, and efficient token caching. The system also features schema specialization for MCP calls, adapting to model size and retrieval settings. Foundry IQ knowledge bases demonstrate superior cost-recall tradeoffs, operating on the efficiency frontier, and consistently improve delegation accuracy and answer quality across diverse datasets. For instance, they lower the no-answer rate by 94.5% and increase evidence recall by 37.9% compared to BM25 search. The functionality is available in the latest preview API.

Key takeaway

For AI Engineers optimizing retrieval-augmented generation (RAG) systems, Foundry IQ knowledge bases present a compelling alternative to standalone search tools. You should consider integrating these knowledge bases to achieve up to 54% higher evidence recall and 34% token cost savings. This allows you to deploy more responsive, accurate agents by leveraging dynamic retrieval, enhanced semantic ranking, and specialized schema adaptation, ultimately improving answer quality and reducing no-answer rates in enterprise applications.

Key insights

Foundry IQ knowledge bases significantly improve agentic retrieval's evidence recall and answer quality while reducing token costs.

Principles

• Dynamic agentic retrieval loops enhance query customization and follow-up.
• Schema specialization optimizes agent guidance for varying model sizes.
• Ground-truth nugget metrics provide robust RAG evaluation.

Method

Knowledge bases orchestrate agentic retrieval via dynamic loops, custom queries, content review, and follow-up queries, synthesizing grounded answers.

In practice

• Deploy knowledge bases to combine diverse structured and unstructured data sources.
• Tune knowledge base configurations (e.g., gpt-5.4-mini) for specific cost-recall balance.

Topics

• Foundry IQ
• Knowledge Bases
• Agentic Retrieval
• RAG Performance
• LLM Cost Optimization
• Evidence Recall

Best for: Machine Learning Engineer, NLP Engineer, AI Engineer, MLOps Engineer, AI Architect

Related on AIssential

• Foundry IQ as a Shared Knowledge Brain for Multiple AI Agents — Foundry IQ centralizes knowledge for multiple AI agents, eliminating redundant retrieval setups and ensuring consistent responses.
• Answer synthesis in Foundry IQ: Quality metrics across 10,000 queries — Answer synthesis in Foundry IQ delivers grounded, cited responses, enhancing RAG applications with steerable and partially complete answers.
• Grounding Your LLM: A Practical Guide to RAG for Enterprise Knowledge Bases — RAG systems combine LLMs with dynamic knowledge retrieval to provide accurate, auditable, and updatable answers for enterprise data.
• How Agentic RAG Works? — Agentic RAG transforms linear retrieval into a decision-making loop, enhancing accuracy for complex queries.
• The RAG era is ending for agentic AI — a new compilation-stage knowledge layer is what comes next — Agentic AI requires pre-compiled, task-specific knowledge artifacts, moving beyond traditional RAG's inference-time reasoning.
• Scaling AI Knowledge Systems: Lessons from the DHTMLX MCP Server — Scaling RAG systems for multiple distinct products requires intelligent query routing and optimized knowledge structuring.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.