Jev Is Overhyped, but I’m Keeping It
Two days after adding Jev as a decision judge: a typed integration and low cost, but an LLM judge was closer to my own job-fit assessments.
Notes from the work
Writing
Thoughts on engineering leadership, software architecture, teams, and applied AI.
2026
Calibrating an AI Job Classifier with Human Judgement
What a small comparison between Jev, an independent LLM judge and my own assessments revealed about job-fit classification.
A Job Title Is Not a Job Fit
Why I am using Jev as a second-stage reviewer for senior technology roles, comparing opportunities with the kind of work I actually want.
My Raspberry Pi Runs the Agent, Not the Model
How I use a 4 GB Raspberry Pi as an always-on home for Hermes without asking it to run the language model.
What 1,031 LinkedIn Job Listings Were Hiding
What 1,031 UK engineering leadership listings revealed about working patterns, changing responsibilities and the limits of comparing job titles.
The Astro upgrade that removed all my spacing
What I learned upgrading this site from Astro 5 to Astro 7, including content collections, Tailwind CSS 4, and the CSS bug that passed every check.