Efficient ML
Training and inference experiments, optimizer variants, quantization, and small models that are easier to understand than they are to hype.
I build and break things around machine learning systems, GPU kernels, local-first tools, and agent workflows. This is a small home base for experiments that deserve a URL.
Training and inference experiments, optimizer variants, quantization, and small models that are easier to understand than they are to hype.
CUDA, performance profiling, memory behavior, and the satisfying feeling of deleting a bottleneck instead of decorating it.
Tool-using coding agents, durable experiment workflows, browser automation, and software that earns its keep instead of adding ceremony.