I built an API that matches resumes against job postings with AI. One simple rule drove the whole design: the LLM is expensive, embeddings are cheap. The result is an API where the LLM shows up in exactly three places, and in none of them does it decide the number that matters.
I finished the courses for the Claude Certified Developer – Foundations certification knowing less than I thought. So I built an app to study with, asked Claude to help me write and correct a question bank, and practised until exam day. I passed by the narrowest margin, and that says something too.
I was about to build a feature that inferred the priority of my open items with an on-device model. Before writing it I measured it against 67 items I had labelled by hand myself. The agreement with my own judgement was zero.