SoftHebb: The Softmax Variant Behind Backprop-Free Learning
How SoftHebb, a small tweak to the softmax equation paired with a local update rule, lets neural networks learn without backpropagation, and where it currently falls short
Runbook
Technical posts about machine learning and data engineering ...
Latest posts
The Lottery Ticket Hypothesis
AI research project exploring the lottery ticket hypothesis
Dispatch from the Frontier - Adventures in AI Engineering
AI engineering a greenfield project (gitgym) with Claude Code
Why Pre-Commit Hooks Are Essential (Even If You Hate Them At First)
pre-commit tutorial
Notes on ripgrep and fd
Notes on ripgrep and fd terminal CLI tools
Exploring Feature Stores with Apache Iceberg
Tutorial on features stores using Apache Iceberg and Python
Precision & Recall
Recall and precision classification metrics tutorial
Social Media Bots
Social media (Bluesky) bot project in Python
My Most Useful VSCode Shortcuts
VSCode productivity improvement with shortcuts, snippets, and extensions
Measuring Sycophancy in Large Language Models
Explainer article about Sycophancy in AI models and how to measure it