Writing
This section is designed to grow into the center of the site. For now, these are short previews of subjects I plan to document in depth.
AI-PEAT started as a question about orchestration: how can a user move from a biological target to useful computational analyses without manually operating every tool and cluster job? The interesting work was not simply calling an LLM. It was defining reliable state, parsing requests, submitting dependent Slurm jobs, tracking outputs, and presenting artifacts coherently.
A small thread pool forces several concepts to meet in one place: ownership, condition variables, work queues, packaged tasks, futures, exception propagation, and orderly shutdown. The implementation is compact, but getting its lifecycle correct is more educational than merely making tasks run in parallel.
Quantitative models turn abstract probability, differential equations, and linear algebra into programs whose behavior can be measured. They expose the consequences of discretization, numerical stability, model assumptions, and computational cost.