Blogs

Thoughts, notes, and experiments.

Batch-Invariant Kernel Design: Conv2d and Attention BMM
Designing fixed-reduction CUDA paths for VLA vision patches and attention.
Aug 2026
Batch-Invariant VLAs
Tracing numerical divergence through SigLIP, dispatch, and batched matrix multiplication.
Aug 2026
Reconstructive Memory for Learning Systems (v0)
Cognitive Science - based framing of the memory problem for learning systems.
Jan 2026
Hello World
A sample post to get things started.
Oct 2025