~/blog / methodology
Methodology
3 posts tagged Methodology.
-
Three Ways to Fake Wi-Fi, One Merged: Zephyr's New hwsim Linux has had mac80211_hwsim for years; Zephyr had nothing like it — you couldn't test the Wi-Fi supplicant without a radio on the bench. Three different fixes landed on the table at once; an RFC and a Wi-Fi telco sorted out which does what, and I built and merged the in-process one: N virtual radios in a single image, running real WPA2 and WPA3 handshakes with zero hardware. Here's the design call, the one problem that makes it hard on Zephyr specifically, and the war story of getting SAE green. -
Same Bug, Two Models: One Wasted a Day, One Found the Silicon Limit Two real nRF Wi-Fi bugs debugged with AI agents. On the same bug, a cheap model burned a day on confident fiction while a frontier model found a hardware limit in minutes — and a fix that passed on the board was still wrong. A field report on model choice, the symptom lying about its layer, and the instruments — sniffer, profiler, bisect — that actually settle it. -
The Dotstar verification loop: from pre-silicon to cloud — no shortcuts How we combine agentic AI acceleration, hardware-in-the-loop testing, wireless performance suites, and senior architectural governance for IoT that survives reality — not just the lab.