/EdgeLog · About
ecro
My name on it. A craftsman's discipline, out on the frontier.
- Record
- 3 products, made alone Embedded + AI + RoboticsA craftsman of small, deep technology
How I work
I have worked in embedded for 18 years, and I now build AI and robotics on top of it. I take an idea to a product that actually runs, and carry it alone from development through certification, production, sales and support.
I would rather build a small brand that lasts than a company that grows fast. I ship few products and keep raising the finish on each one. They do not need to sell to everyone. A few people trusting one for years is enough.
I look at the problem before the technology. Before deciding what to build, I ask what changes for the person using it once it is solved. A boot that is ten seconds faster matters less than one fewer stall in someone's day.
EdgeLog is where that work gets written down: what I picked and what I dropped, the runs that failed, and the part numbers behind them.
Operating rules
- I build what I need first. Every project here started as a tool I was missing.
- Small experiments, then one thing held for long. A new idea gets a prototype first. I only dig deep into what keeps pulling me back.
- Few products, well finished. I cut scope to the smallest thing that keeps the promise. I don't cut quality.
- Measurements decide. A hypothesis, a measurement, then keep, change or stop. I set what a failure may cost before I start.
- AI for what one person couldn't do. Faster is not where it ends. It does not take over the user's judgment.
- My name stays on it. What ships keeps getting fixed, and when something is fixed, I say so.
If something I made helped you, tell me. Those notes decide what I fix next.
Work
-
Tuned in development
Tennis impact sensor, Tuned Dot, on nRF54L15 and Zephyr, with a Flutter app.
Zephyr · C · Flutter
Get notified → -
NeuroTerm maintained
Serial and SSH terminal for embedded developers who work with coding agents. Terminal Use lets the agent read and type on your board; Work memory keeps each board's record.
Tauri 2 · React · Rust
neuroterm.dev → -
harness-maker maintained
Project-shaped agent harness. On PyPI, grade-gated review.
Python · Jinja2 · PyPI
GitHub →
Research
- EmbedEval LLM benchmark for embedded firmware. 267 cases, 5-layer eval. Python · Open Source
Three axes
The record is kept along three axes. Products change; the axes stay.
- Making
- The story of each product, one decision at a time: what I chose, what it cost, what happened next.
- Methods
- Ways of working that carry over to any project, measured: AI agents, firmware, review loops, and the runs that failed.
- Philosophy
- A record of looking for a philosophy I don't fully know yet. It covers several layers: my own, my products', and more.
Contact
Using something I made, or making a product alone too? Write. I'm glad to trade notes on anything.
contact@edgelog.dev