HDNA Workbench: open-box AI research platform where every neuron is inspectable with one-line PyTorch inspection wrapper [P]

About this article

GitHub: https://github.com/staffman76/HDNA-Workbench I built this because I was tired of studying AI through black boxes. Existing interpretability tools explain models after the fact. I wanted an architecture where transparency is the design, not an afterthought. What it does Open-box AI engine (numpy only): - Neurons with persistent memory — query any neuron's activation history, routing, and health at any time - Pluggable reasoning daemons with quality-gated phase progression - Audit log t...