R&D BAY
Lab
The home laboratory - where 25+ years of ops discipline meets local machine learning. Everything on this bench gets measured, versioned, and documented, because "it works on my GPU" deserves the same rigor as production.
Local LLMs
Running open-weight models on local hardware: inference servers, quantization trade-offs, context-length limits - and what it actually takes to self-host a capable assistant.
GPU Benchmarks
Measuring what local hardware really delivers: tokens per second, VRAM ceilings, batch sizes, thermals under sustained load. Numbers, not vibes.
Experiments
Fine-tuning small models, RAG pipelines, and testing what transfers from Kaggle notebooks to the home rack - with the failures logged alongside the wins.
// write-ups land in notes and on the blog as the numbers stabilize