Three independent regressions in OpenNutriTracker converged on one wrong daily calorie number. Each was hard to spot alone. Diagnosis felt like peeling layers.

Two free public APIs power every food lookup in OpenNutriTracker. Both are flaky enough that retry, cache, and custom-meal fallback need to be a layered system.

How I built Cinnamon panel applets for two wireless peripherals that don't expose battery anywhere on Linux: the Logitech A20 X headset and Yunzii B87 keyboard.

QR codes can hold more data than you'd think. But 'technically fits' and 'scans reliably on a real phone' are different questions, and the gap drove every interesting decision.

A follow-on to Taking Up Space: the internal mechanics of imposter syndrome at senior level, and the difference between calibrated doubt and self-erasure.

From a personal workaround to open source release: what the gap between 'works for me' and 'works for anyone' looks like, and what I'd do differently.

Treating prompts like a CLI spec rather than a search query: the mental model shift that actually changed how much I get from AI tools.

Not a review of Claude Code's features but an account of what concretely changed in my day-to-day work after six weeks of using it seriously.

Wiring Azure private endpoints to central DNS zones via a structured Terraform output pattern: the approach, the pipeline contract, and the gotchas.

From someone who uses both professionally: a comparison focused on real friction rather than feature matrices, and when each one earns its place.