Topics

A running table of contents, my archive of learning and occasional write ups.

  1. Browser Automation exploring

    Driving a real web browser programmatically for testing, scraping, and increasingly for AI agents that navigate the web.

  2. C4 Architecture exploring

    A lightweight model for visualizing software architecture at four zoom levels — Context, Containers, Components, and Code — so diagrams stay consistent and easy to follow.

  3. Verifying correctness empirically through testing, simulation, and coverage, the practical counterpart to formal (proof-based) verification.

  4. CUDA Programming exploring

    Programming NVIDIA GPUs directly: the execution model, the memory hierarchy, and how to write and launch kernels.

  5. Diffusion Models exploring

    Generative models that create data by learning to reverse a gradual noising process, behind most modern image generation and now text.

  6. Edge Models exploring

    Running models on-device (phones, laptops, embedded) under tight memory, latency, and power budgets.

  7. Finetuning exploring

    Adapting a pretrained model to a specific task or domain, from full finetuning to lightweight LoRA.

  8. Proving software and hardware correct against a specification, instead of only testing for bugs.

  9. Kernel Engineering exploring

    Writing and optimizing the small, hot compute kernels (often on GPUs) that make ML and numerical workloads fast.

  10. Reverse-engineering what neural networks actually compute, and the tooling (like Goodfire) that makes internals inspectable and steerable.

  11. Open Source Models exploring

    Openly available model weights (Llama, Mistral, Qwen, Gemma, DeepSeek) and the licenses and ecosystems around them.

  12. Rust Programming exploring

    A systems language that guarantees memory safety without a garbage collector, through ownership and borrowing.

  13. Shor's Algorithm exploring

    A quantum algorithm that factors large integers in polynomial time, threatening the RSA encryption that secures much of the internet.

  14. Compact language models (often a few billion parameters or fewer) that trade scale for speed, low cost, and on-device use, while staying surprisingly capable.