Benchmarking Microsoft's Knowledge Layer: Foundry IQ multi-hop RAG pipeline dissected
adnan_hashmi · x · 2026-10-25
A measured look at Microsoft's Knowledge Layer built on Foundry IQ and Azure AI Search. The pipeline: gpt-4.1 query planning splits questions into 1-5 subqueries, runs parallel hybrid search with semantic reranking (up to 50 chunks each, 108 indexed), merges and ranks into a cited grounding payload, then branches into extract (RAG prompt answered by gpt-4.1) and synth (knowledge base writes the answer) paths, evaluated by a judge. A Fabric IQ ontology with entities, relationships and rules serves as a remote knowledge source.
More from coding & agent
- Adaptive Skill Granularity Framework Routes Coding Agent Skills at Runtime, Hits 100% Success — dair_ai · 2026-10-25
- Study: LLMs rewriting their own ARC-AGI-3 harnesss gain just 2.8 points, not significant — omarsar0 · 2026-10-25
- Dev adds thread tags to agent sidebar to track orchestrators across machines — kevinkern · 2026-10-25
- Dev experience: once your agent works well enough, you can ask it how to improve itself — Then-Programmer7221 · 2026-10-25
- AIOps didn't cut alert volume—the bottleneck is the runbook no one trusts to automate — krishnan · 2026-10-25
- Hermes ships subagent routes plugin for per-task model orchestration — intellectronica · 2026-10-25