Writing

On AI engineering, privacy, cloud infrastructure, and building systems that last. RSS ↗

The data you didn't mean to train on

Why DSR workflows break the moment personal data enters a training pipeline — and what to build before that happens.

privacy · ai 2026-08-12

Build the harness before you need it

The same reason you backtest a trading strategy before putting money behind it applies to LLMs in production.

ai · infra 2026-07-14

Evals are not tests. That distinction costs teams months.

The gap between "correct" and "acceptable" is where most teams lose time they didn't know they were spending.

ai · evals 2025-06-01

Privacy Engineering Habits for Backend Teams

Practical habits that make access and deletion flows less painful when regulations eventually show up.

Field Notes · 2019–2021

Older technical notes from past projects. Published here as a record, not as current recommendations.

Notes on Asynchronous File Watching

inotify, FSEvents, and the edge cases that make file watching harder than it looks in trigger-based ETL.

utilities 2021-05-01

Container-First Modernization — A Lightweight Playbook

What actually works when containerizing a legacy ETL stack without a six-month roadmap.

platform 2020-03-01

Building a Simple Strategy Tester in Python

A minimal backtesting harness that teaches you where strategies actually break.

algo trading 2019-11-13

How to Test a Trend-Following Trading Strategy

Statistical rigour for strategies that look good on paper and need to prove it.