Notes · 10 posts

Things I've learned the hard way.

Writing on data engineering, AI in production, and migrations that didn't go to plan. The fuller versions of the work behind the case studies.

Jul 8

Primitives, not features: why Pi is my go-to agent harness

Pi is a deliberately minimal agent harness. That restraint is why it is now my default for personal and client work, and why it saves so many tokens.

AIagentsopen-source
Jul 2

Hermes Agent: a local, open agent built around skills

Nous Research's Hermes Agent is open source, local-first, and model-agnostic, and it turns your own workflows into reusable skills. Why that design matters.

AIagentsopen-source
Jun 25

The bottleneck wasn't the warehouse: cleaning legacy master data with a local SLM

A legacy warehouse's reports were wrong because the master data underneath was dirty. How a local, air-gapped SLM cleaned it in place, with no data leaving.

data-qualitydata-migrationgdpr
Jun 18

Own the model, not the API: cleaning SAP master data with a local SLM

SAP master data can't go to a cloud LLM in a regulated shop. So I fine-tune a small model you own: air-gapped, audited, trained only on synthetic data.

open-sourcegdprdata-quality
Apr 27

3 things that break when you move Databricks notebooks to dbt

The three SQL syntax, schema reference, and state management problems I keep seeing when teams migrate from Databricks to dbt; and how to fix each one.

dbtdatabricksdata-engineering
Apr 20

GDPR-aligned RAG: a checklist that survived three audits

A GDPR checklist for RAG that survived three DACH audits, and the sovereign build that makes them easy: local models, EU weights, data that never leaves.

AIGDPRRAG
Mar 21

Why I built ECL: synthesis over retrieval for enterprise knowledge

RAG returns the closest document. ECL returns the reasoning your experts use, cited and conflict-aware. Open source, built on Andy Chen's pattern.

context-engineeringopen-sourceenterprise
Nov 22

Four failure modes I've seen in enterprise lakehouses (and the cheap fixes)

Four patterns that derail lakehouses in production: the swamp, the performance mirage, the metadata ghost town, and the tables nobody maintains.

lakehousedata-platformfailure-modes
Aug 10

When not to migrate off your on-prem SQL Server

Four situations where keeping SQL Server is the right call, and the cost math the warehouse vendors won't show you.

sql-serverdata-migrationcost
Apr 27

Lakehouse migration: from Databricks to Snowflake for a European media company

How I migrated 20M+ records from Databricks to Snowflake, cutting monthly infrastructure costs by 40% and fixing years of accumulated governance debt.

data-migrationsnowflakedatabricks

Recognize one of these problems?

If a migration, a lakehouse, or an AI system is keeping you up, a quick scan is usually the fastest way to a clear answer.

See the packages → Book a scoping call