Start here
Eleven posts, four themes. If something here is useful, the measurements are the point — I publish the numbers I actually got, including the ones that proved me wrong.
Benchmarks and numbers
Everything starts with a measurement, usually because the vendor's number and my number disagreed.
- I'm betting local AI hardware appreciates faster than gold
- Qwen3.8-Flash-Next on my DGX Spark: every attempt, and what each one bought
- Qwen3.8-Flash-Next on my DGX Spark: tuning decode speed
- Nvidia P40 Total Guide and Lessons Learned
- Intel iGPU Passthrough to LXC: Encode Benchmark
AI agents running on my own hardware
- I Gave My Homelab Agent Long-Term Memory — Here's What Happened
- I ran an AI pentest on my homelab — and misread my own results
- What Paperclip gets right, and why I still run simple-lcp
Guides and how-to
- How to Host Multiple Apps on a Single IP
- I Reverse-Engineered My AC Because the App's Scheduler Was Broken
Building product
There is no email list yet — RSS is the only way to follow along right now.