Blogs and Wiki

Training a base non-recurrent transformer model using self attention to be used as a foundation model from first principles. My base is generative pre-trained transformers in the model of the "Attention Is All You Need" paper 1706.03762 from Google

General CS code - a refresh from my Carleton CS days in the 90's all the way to Stanford in 2025

gcp-infrastructure-as-code

Systematically determine the optimized spot where we can push our hardware to the fullest possible use. Multithreaded optimization depends on multiple factors including CPU/GPU type (M4Max vs 14900 or MetalCUDA. Operations involving space-time tradeoffs like heap usage need to be fine tuned around batch sizes.

Container with a model and inference code

Doppler Radar ML

Biometric Backend in Java/Spring

biometric http endpoint in gcp cloud functions

Generative AII/ML under quarantine - sandbox to test deployments - no generative code in my other orgs