Tech claims, tested.
Benchmarks, build logs, and side projects. Tested myself, not repeated from a video.
James
Partway through an MSc in Artificial Intelligence at the University of Bath. Everything on this site is real work, built and tested on my own hardware: benchmarks, tools, and from-scratch builds like the Enigma machine below.
Connect on LinkedInLatest
Three Local Models, One Bug: Two Fixed It, One Never Answered
I gave three local coding models the same buggy function. Two fixed both bugs in under a minute. The third burned through 16,000 tokens of reasoning and never produced an answer at all.
Fact-Checking a Viral AI YouTuber's GPU Advice
A popular guide told viewers exactly which model to run on a 16GB card. I ran the actual benchmark instead of taking his word for it. His pick lost by 8 points and ran 3x slower than a model already on my disk.
1.34 Million Tokens That Never Touched Claude's Context
I built a tool to route bulk file reads to a local model instead of Claude's context window. Here's what its own call log actually shows, measured against real usage, not a marketing claim.
