Things I'm building.
Working systems, half-finished experiments, and the occasional rabbit hole. If something looks useful, fork it.
Foundry × TOGAF Agent Crew
The hardest part of agentic AI isn't the models — it's deciding who does what, in what order, with what authority. A reference crew that maps Business Architect (market) → Enterprise Architect (GO/PIVOT/NO-GO) → Solution Engineer (MVP) → Implementation Manager (delivery) → Stakeholder Lead (GTM). Best-fit model per role, three orchestration patterns, industry-agnostic.
Power Platform Telemetry → Fabric Lab
End-to-end telemetry pipeline from Power Platform into Microsoft Fabric — a reference lab for capturing app/flow signals, landing them in OneLake, and modeling them for analytics and AI-ready datasets.
Fabric Data Agent Patterns
Enterprise-architecture reference blueprint for evolving a Fabric Data Agent MVP into a Baseline-to-Target architecture. Synthetic CPG data → OneLake Delta → DirectLake semantic model as the semantic contract → BPA conformance → Fabric Data Agent (Assistants v2, grounded on the model, no RAG) → CI eval harness with ungrounded-deflection guard. Six vertical SBBs (CPG, FSI, healthcare, energy, public sector, telco). Responsible AI, FinOps, Zero-Trust, and governance woven in.
AI Trends Weekly
Two AI-generated weekly briefs (grounded + strategic) shipped via Azure Container Apps Jobs to a private list. Foundry agent + Bing grounding + ACS email.