Measured: Microsoft Foundry IQ and the Fabric IQ Ontology, with latency and cost details
adnan_hashmi · x · 2026-10-29
A hands-on measurement of Microsoft's Foundry IQ and Fabric IQ Ontology with an architecture diagram: the ontology agent reads tables, entity types, relationships and rules via the user's token (median 24s, up to 86s); query planning reads the ontology wiki at $1.0 per question across both efforts; the knowledge base runs on Azure AI Search, with HTTP 502 when the ontology can't answer. Synth/extract paths plus a gpt-4.1 judge close the loop.