About

Notes from hands-on work across AI hardware and software: local LLM deployment, inference acceleration, GPU and domestic accelerator selection, multi-node cluster interconnect, and inference on edge devices.

Most posts are benchmark-driven — configs, parameters, throughput and latency numbers come from real environments, with reproduction steps and the pitfalls I hit along the way. Vendor comparisons stay within what I can verify, and anything uncertain is labelled as such.

This site is built as static files: no database, no admin entry point. Posts are written in Markdown and published from the build output.

Contact

Follow updates via RSS.