the blog

Cracking AI.

Practical AI, explained three ways: the intuition, the understanding behind it, and an implementation you can run. Deep dives come as series, read in order; everything else is a standalone article.

10posts 2series 0standalone rss feed
01

Series

Multi-part deep dives. Each one is a single track you read from part 01 onwards.

S02 5 parts 91 min total Intermediate complete

Decision Models (System One) from Scratch

A fast, intuitive model that scores options instead of writing text. Frozen Qwen3-8B embeddings, two small heads, contrastive learning and a three-stage training recipe, explained with the maths and rebuilt end to end.

  1. 01 What Is a Decision Model (System One)? 13 min
  2. 02 Fine-Tuning a Frozen LLM: Why We Train Heads Instead of LoRA 19 min
  3. 03 Contrastive Learning: InfoNCE, Temperature and Hard Negatives 17 min
  4. 04 Training a Decision Model: The Data, and Pre-, Mid- and Post-Training 25 min
  5. 05 Does It Work? A Verifier for Coding Agents, the Results, and What We Learned 17 min
02

Articles

Standalone pieces: paper notes, opinions and experiments.

No standalone articles yet. The series above is the place to start.

03

Archive

Every post, by year.