/About · operating note · v2 · Jul 2026
ecro
Eighteen years in embedded. I run AI agents against real hardware and publish what the measurements say.
18 Years embedded
267 Benchmark cases
03 Certified releases
What gets written
EdgeLog is the record of designing products, building them, and breaking them.
Every post starts from a measurement taken on a real system. I write about what decides whether a product ships, not about a demo that runs well.
Reliability, observability, OTA, security, fleet ops. When a run fails, the post says what failed, down to the part number where that matters.
Four projects sit behind the writing. If the projects are the instruments, these posts are closer to the readings taken off them.
Instruments · four projects
EmbedEval LLM benchmark for embedded firmware. 267 cases across 24 categories, 5-layer eval (L0 to L4). Open source. Python · Open Source harness-maker Project-shaped harness for Claude Code, Cursor, Codex. 12+ stack signals plus a 10-dim interview shape every render. Python · Jinja2 · PyPI
SpotOn in progress nRF54L15 + ICM-42688-P at 1kHz on Zephyr. Modal impact features run on the MCU; the model is still waiting on data. Zephyr · C · Flutter
NeuroTerm Serial and SSH terminal for embedded engineers. Local LLM, offline RAG, pattern markers. Tauri 2 · React · Rust