Writing

Writing

Essays on data infrastructure, AI deployment, and the practice of consulting.

01

Why most enterprise AI fails in the boring middle

The problem is rarely the model. It is the twelve steps between the model and a decision that someone will act on.

Data Engineering · 2026 · 8 minute read8 minute read
02

The economics of a data pipeline rebuild

When a client asks how long a rebuild will take, the honest answer starts with measuring what the current system actually costs.

Infrastructure · 2025 · 11 minute read11 minute read
03

On anonymization: what you can and cannot hide

We have anonymized client data for five years. Here is what we have learned about what is genuinely safe and what only appears safe.

Practice Note · 2025 · 6 minute read6 minute read
04

The pipeline failure modes no one talks about

Silent failures are the dangerous ones. Here are the four failure patterns we encounter most often, and how we instrument for them.

Infrastructure · 2024 · 9 minute read9 minute read
05

What "production-ready" actually means for an ML model

It is not about accuracy. It is about the eleven properties a model must have before you can safely operate it.

AI Systems · 2024 · 13 minute read13 minute read