Field Log

Chapters from the build log — write-ups on AI/ML experiments, systems I've shipped, and whatever rabbit hole I fell into that week.

Shaving Milliseconds Off LoRA Inference for Diffusion Models

Notes from swapping adapters at inference time without eating a full UNet forward pass on every merge.

diffusion-modelslorainferencepytorch

Building a RAG Pipeline That Doesn't Fall Over in Production

The retrieval half of RAG breaks in ways that don't show up in a demo, and here's what caught the failures in mine.

ragllmagentsvector-search

Down the Rabbit Hole: What Writing a Toy Autograd Engine Taught Me

I called PyTorch's .backward() a thousand times before I understood what it was actually doing, so I built a tiny version to find out.

deep-learninglearning-in-publicpytorch