Blog

Operational Data Quality Notes: separating incident response from root-cause fixes

Agent Workflows in Practice: state handling patterns that reduce agent confusion

RAG Systems That Hold Up: using reranking only where it changes outcomes

LLM Reliability in Practice: treating quality as a product metric

Microsoft Foundry Build Notes: moving from model demos to governed operations

Real-Time Signals That Actually Help: building alerting that avoids fatigue

SQL in Fabric: Working Notes: building models analysts can trust without constant hand-holding

Data Movement Without Drama: designing pipelines for failure, not the happy path

Keeping OneLake Clean Under Delivery Pressure: balancing speed and access boundaries

Fabric Architecture Notes: turning messy raw zones into reliable products

Cost Discipline for LLM Apps: reducing token waste without hurting answer quality

Data Quality Work That Actually Sticks: choosing the minimum useful set of quality signals

Building Useful AI Agents: tool access policies that improve reliability

RAG Tradeoffs I Keep Seeing: fixing retrieval before touching prompts

How I Evaluate LLM Changes: tracking groundedness before celebrating fluency

Foundry Decisions I Stand Behind: how project boundaries change delivery speed

Real-Time Signals That Actually Help: keeping streaming dashboards useful after week one

Fabric Warehouse Tradeoffs: choosing model grain before performance tuning

Pipelines I Trust in Fabric: reducing brittle dependencies in daily loads

Keeping OneLake Clean Under Delivery Pressure: why governance has to be designed before scale

Lakehouse Decisions I Made This Week: turning messy raw zones into reliable products

The Invisible Tax: Why the Future of AI Is About Transitions, Not Tasks

The highest-cost gap in knowledge work is the transition between commitments. Transition intelligence is where proactive AI can create compounding value.

March: Fixed Sleep, Fewer Tabs, More Focus

February 2026: Month in Review