7 min read
A Token Waste Audit: Cut What the Answer Doesn't Use
Profile each LLM request by segment, rank the cuts by quality risk, and keep only the token savings that survive a paired evaluation against the baseline.
Blog
Thoughts on data engineering, cloud architecture, and technology. 2205 articles and counting.
Showing 24 of 2205 posts — page 5 of 92
7 min read
Five Data Quality Signals for the Data-to-AI Handoff
Why I'd track five quality signals per feature table at the data engineering to AI handoff, and how to compute them in a Fabric notebook without a new tool.
6 min read
Least-Privilege Tools for AI Agents: Four Layers of Control
Agent reliability depends on which tools a model can reach. Four layers of tool policy (surface, schema, approval, budget) with Agent Framework code.
7 min read
Wrong RAG Answer? Rule Out Retrieval Before You Edit the Prompt
Most wrong RAG answers start with the passages, not the prompt. A labelled question set and a retrieval-only triage on Azure AI Search show which one to fix.
8 min read
Gating LLM Changes on Groundedness Flips, Not Fluency Averages
How I compare a baseline and a candidate prompt or model on frozen context, count groundedness regressions per row, and treat fluency as a tie-breaker.
8 min read
Foundry Project Boundaries: Draw Them Around Use Cases, Not Demos
How I split Microsoft Foundry resources and projects by environment and use case so RBAC, quota and agent data line up with ownership and delivery speeds up.
8 min read
Why Real-Time Dashboards Rot After Week One, and How to Stop It
Rules for Fabric Real-Time Dashboards that stay trusted: tiles with owners, visible data freshness, refresh costs you can afford, and thresholds in tables.
7 min read
Fabric Warehouse: Decide Fact Table Grain Before You Tune Anything
Why fact table grain, not clustering or statistics, sets the performance ceiling in Fabric Data Warehouse, and how to declare and test it in T-SQL.
9 min read
Fabric Daily Loads: Gate on Data Readiness, Not the Clock
How to remove brittle dependencies from daily Fabric loads with one orchestrating pipeline, readiness checks, idempotent writes and deliberate retries.
8 min read
OneLake Table Contracts: Owners, Schemas and Failure Modes Up Front
Why OneLake governance has to be designed before scale: table contracts, schema-enabled lakehouses, explicit owners and deliberate failure modes in Fabric.
9 min read
Table Contracts in a Fabric Lakehouse: Where Raw Becomes a Product
Four decisions for a messy Fabric raw zone: contracts at silver, Delta constraints, quarantined rows with failed batches, and lakehouse schemas.
9 min read
Transitions, Not Tasks: Designing AI for the Gap Between Meetings
Most AI assistants help inside a task. The expensive part of knowledge work is the switch between commitments, and that needs a different agent design.
5 min read
March 2026: A Fixed Bedtime, Two Inbox Checks, One Tool to Ship
March plan: turn February's two drifting goals, sleep and email, into fixed rules, keep what held, and ship the side project I've already built.
7 min read
February 2026 Scorecard: Four Commitments, One Short Month
Grading my four February commitments: 29 posts in 28 days, two no-tech Sundays, one new engagement against two noes, and sleep stuck at 6.5 hours.
5 min read
Do I Miss Being a Principal Engineer? Yes and No
Someone asked if I miss being a principal engineer. What I miss, what I don't, and how to tell good reasons for staying technical from bad ones.
8 min read
Equipment Monitoring on Fabric Real-Time Intelligence, End to End
One sensor-monitoring design on Fabric Real-Time Intelligence: where to filter, enrich, aggregate and alert, and what Eventhouse retention costs you.
5 min read
Fifty-Five Days In: Being Wrong in Public and Writing on Empty
Fifty-five days of daily posts: calibrating my opinions by being wrong in public, letting go of saved-up ideas, and why the empty-tank days are worth it.
5 min read
The Code Review Comment That Wasn't About the Code
Seven years ago a tech lead answered my pull request with one question about who it was for, and it still shapes how I review code.
9 min read
Azure AI Foundry Is Now Microsoft Foundry: Hubs, Projects, Choices
What the Microsoft Foundry rename means in early 2026: Foundry projects vs hubs, the preview portal, prompt flow's place, and how I'd decide.
8 min read
KQL for SQL People: The Query Language Azure Already Runs On
Why KQL is a baseline Azure skill, how its pipe model differs from SQL, and the operators, join defaults and has-vs-contains traps that matter most.
4 min read
Second No-Tech Sunday: Two Phone Reaches and Better Questions
Our second family day without screens: fewer phone reaches, kids who didn't argue, a sharp observation from my wife, and the work anxiety that still shows up.
8 min read
Fine-Tuning or RAG? Decide by the Failure You're Fixing
Fine-tuning and RAG fix different failures. How to tell which one you have on Microsoft Foundry in early 2026, what each costs, and when to combine them.
6 min read
Last Known Good: A Debugging Lesson From My Son's LEGO Set
My nine-year-old fixed a stuck LEGO build by going back to the last step that worked. It's the debugging discipline I'd let slip on a Fabric pipeline.
7 min read
Delta Lake in Fabric: Not Every Engine Reads Every Table
Every Fabric engine speaks Delta Lake, but not the same features. How to use the interoperability matrix so Spark tables still work in SQL and Power BI.