Engineering notes.
Written while doing the work rather than for a content calendar. Each one answers a question we were asked often enough to want a permanent answer.
-
Designing IoT for the real world: unreliable networks, OTA, and data you can trust
Field devices live on flaky networks where readings get lost or duplicated. Here is how store-and-forward firmware, idempotent ingestion, and managed over-the-air updates produce telemetry you can actually act on.
-
Cut your cloud bill without cutting reliability: when self-hosting actually pays off
Most cloud overspend is idle capacity, not safety margin. Here is how to right-size a stack onto self-hosted containers, and the honest cases where managed services are still the correct choice.
-
Getting cited by AI: a practical GEO playbook for technical sites
Generative Engine Optimization is about being quoted in AI answers, not just ranked in search. Here is the structure that makes a page extractable, trustworthy, and citable by ChatGPT, Perplexity, and Gemini.
-
Core Web Vitals in 2026: earn 100/100 by construction, not by patching
Most Core Web Vitals work is undoing damage a heavy framework already did. Build static-first and the scores come for free. Here is the architecture that makes 100/100 the default.
-
Reliable LLM pipelines: let the model propose, let the schema decide
LLMs are excellent at reading messy human input and unreliable as a source of truth. The fix is an architecture where the model extracts and a strict schema validates, with retries and human review.
-
Why 3-D Secure payments fail intermittently, and how to fix the signature
Intermittent hash-mismatch declines under 3-D Secure almost always trace back to non-deterministic request signing. Here is how to make signatures canonical and stop the silent failures.