I write about the tools that actually ship.
Staff engineer · ML accelerator hardware · kernel & firmware
Who I am
I'm a staff engineer at a large semiconductor company, where I work on ML accelerator hardware — writing kernels, firmware-level code, and analysis scripts for data-driven decision-making across device fleets. My day-to-day stack is C, C++, Bash/Shell, and Python.
I've been using AI tools in real engineering workflows since the early days, and I've watched the space evolve from novelty to infrastructure. I run my own home server, deploy everything in Docker containers, and spend a lot of free time experimenting with local LLM setups.
Tools I actually use
AI / LLM
Claude Code, OpenClaw, Claw Code — and local deployments via Ollama and llama.cpp. I've experimented broadly across models and I know what each one is actually good at.
Dev stack
C, C++, Python, Bash. Docker for everything self-hosted. I've built and maintained production systems at scale — not just side projects.
Photography
Sony a7C II. Tamron 70-180mm for reach. Sony 40mm f/2.5 G for street — small, fast, disappears in a crowd.
Outside the terminal
Built a racing drone from scratch with Betaflight. BBQ obsessive — grill, smoke, wood, coal, all of it. Currently running an Ooni Karu 12G for pizza.
Why this blog
I got tired of AI tool reviews written by people who've never shipped production code. The "I tried this for a week" takes miss everything that matters: how tools hold up under real constraints, what breaks at scale, where the gap is between the demo and the job.
This blog is my notes. What I actually use, how it holds up, and where it falls apart. No fluff.
Sharing the home server rack with an Akbash/Maremma mix, an Australian Shepherd, and a Siberian cat. They are not impressed by the uptime.
Review policy & affiliate disclosure
All reviews on this site are based on personal use. I pay for or independently obtain the tools I write about. Where affiliate links are present, they are disclosed inline on the relevant page. No sponsored content is published without explicit labelling.