Software Engineer & Writer
Building software, writing about the process, and sharing what I learn along the way. Currently exploring React Native, developer tools, and the craft of simple code.
A component-freezing ablation study on modular division reveals that grokking isn't one process. It's two: infrastructure setup, then computational reorganization. The boundary between them is sharp and measurable.
A 3-layer MLP learning F = m*a behaves completely differently with 63 neurons vs 64. The investigation reveals three training regimes, winner-take-all gradient dynamics, and a lesson about what you're actually ablating when you change architecture.
Four independently trained models find the same subspace geometry. That geometry transfers cross-modally, defines what solutions are reachable, and guides developmental construction toward convergent structure. The evidence, the controls, and what it means.
Production-grade LLM inference API built from scratch. NestJS gateway + Python GPU workers. Scheduling, batching, KV cache, tensor parallelism, multi-modal, all against real GPUs.
Figma-compatible Rust/WASM canvas engine. 134M-point rendering, built-in CRDT collaboration, native .fig import — one binary, zero JS dependencies.