About

I work in technology and spend a lot of time thinking about AI, software, learning, and building useful things.

This site is my public notebook: tools I am trying, things I am building, and concepts I am learning.

A low-friction place where I can put thoughts before they become stale.

Recent Articles

  1. LLM training, seen as data

    A visual walkthrough of modern LLM training through the data at each step: pretraining rows, midtraining mixtures, supervised fine-tuning, preferences, RLHF, reasoning RL, tool use, agent trajectories, domain data, safety, distillation, and evals.

    June 22, 2026
  2. Building a tiny browser-playable world model

    A visual walkthrough of a tiny Breakout world model, from simulator data to a playable neural browser demo.

    June 15, 2026
  3. Agent-native gateway

    A production-focused, visual explanation of the agent-native gateway pattern: how coding agents safely discover, authenticate, and call internal APIs, MCP servers, and third-party applications through an open-source enterprise stack.

    June 7, 2026
  4. What training a tiny text-to-image model looks like

    A first-principles visual walkthrough of training a small latent text-to-image generator from random weights: captions, embeddings, tensors, latents, tokens, flow matching, loss, sampling, Self-Flow-lite, monitoring, and prompt coherence.

    June 2, 2026
  5. LLM inference, from request to token

    A visual, source-grounded walkthrough of LLM inference and serving using nano-vllm and Mini-SGLang: tokens, prefill, decode, KV cache, scheduling, batching, prefix caching, CUDA graphs, tensor parallelism, and streaming.

    May 31, 2026
  6. Building a wildfire simulator with neural cellular automata

    A visual walkthrough of cellular automata, neural cellular automata, local prediction, control, and how training would teach the shared rule.

    May 29, 2026
  7. The die is not the whole product

    A visual note on why advanced AI chips are shaped by packaging, memory, heat, alignment, power, cooling, and manufacturing.

    May 24, 2026
  8. Evals with Harbor Framework

    A problem-first walkthrough of building and running a Harbor Framework eval for an agent that repairs a token billing ledger.

    May 24, 2026
  9. Follow the numbers through an AI chip

    A visual explainer about why AI chips are built around data movement, and how that turns into latency, batching, memory bandwidth, and long-context costs.

    May 23, 2026
  10. Ask for the other drafts

    A small note on verbalized sampling and using AI to write more creatively.

    May 2, 2026
  11. This site

    A public notebook with four simple shelves.

    April 25, 2026
  12. AI and software

    Notes on how AI changes the way software gets made and learned.

    April 25, 2026
  13. Low-friction notes

    A reminder to publish small thoughts while they still have energy.

    April 25, 2026
  14. Astro and Cloudflare Pages

    A small test of a plain Markdown-first publishing setup.

    April 25, 2026

Elsewhere

GitHub LinkedIn X / Twitter