Now

There are only 2 more years until the permanent underclass is officially established. Maybe I (and you) will be part of it together!

Basics

Education: CS @ UCSC 2020-2023 (Yes, I completed it in only 3 years)

Work: This (more info to come soon).

Residence: East Bay Area

Open Source

I also contribute to open source, and many of the projects I've done in my free time are also open source. Here are some open source things I'm currently actively maintaining or contributing to:

• Swift Operation (A better Tanstack Query for Swift.)
• Cactus (Low-latency on-device AI inference.)
• Various other Swift libraries that some people find useful. (See my Github profile.)
• I also occassionally chip in to PointFree's libraries.

In Progress/Planned Writings

As opposed to traditional software projects which can be spun up nowadays in a matter of hours (thanks to agents), I prefer to be very methodical and take the scenic route with my writings.

All of my writings for this site are hand-crafted, and no AI is used whatsoever for generating the actual written content.

• Locality of Decisions (In Progress)
• An interactive systems design piece that introduces a conceptual framework around representation, optimization, and meaning (not spiritual, but systemical). Also explores the issues with many of today's mediums, leveraging ML/AI, and dynanicism.

Some Used Tooling

• Languages: Swift, Typescript, C++ (ML forced me), Python (ML forced me)
• Frameworks: MLX, SwiftUI, TCA, React, React Native, SvelteKit, and so on. Ok is this just sounding like the "skills" section of a resume???
• Agents:
  • GPT 5.6 Sol is great value (mostly on medium and high), and my go-to for complex tasks with Codex. I also think Luna and Terra (especially Luna) are very underated, and they are my go-to models for scopeable/low-stakes tasks and automations to save on usage.
  • For Anthropic, I mostly use Fable to drive larger features, and rely on Opus (or even Sonnet sometimes) to handle the scraps. Opus 5 is also good enough to handle most tasks on its own, but I really do not like it's overall personallity when talking to it. (Seems to be a common sentiment, but I'm not one of those people who think Opus 5 is unusable.)
  • I generally prefer Codex on the day to day usage basis (chatting/planning is smoother), but I prefer Claude for tasks that require more analysis (eg. optimizing performance) or UI prototyping (though I think you can get good UI out of any model if you actually take the time to design things).
  • I use Pi for everything at my desktop right now. Codex Desktop and Opencode were my previous daily drivers, but Pi has a larger extensibility culture that I like more.
  • For remote work, I'm using the T3Code mobile app, which should come with a disclaimer that it will infect what little life you have outside being a clanker (eg. Ruining social interactions and sleep schedules). My 8-year old iPad Pro 11' is now a genuinely viable dev machine.
  • I also dabble with openweight and local models from time-to-time like GLM 5.2, Minimax M3, Gemma 4, etc. They tend to be pretty capable for many code-related tasks, but I would still leave larger tasks to the frontier (GLM 5.2 can also work for these).
  • In terms of handwriting code, I usually do so mostly for learning purposes and pure fun now. (ie. This improves your ability to use agents effectively.) The other cases are mostly for "feeling out the right API" when I work on libraries/frameworks, and for correcting minor things inside performance intensive settings. However, I find the latter case becoming less-and-less common as the models improve since I can now mostly get the speed by dicating the higher-level intent. (eg. "We can use SIMD to parse this faster, explore all cobinations of options for this.")

Currently Learning

• Anything related to building ML frameworks.
• World Models seem interesting.
• Psychology, especially around "intrinsic motivation" (via books like this [academic, not self-help]).

Human Info

I happen to be part of the "human" species, here is some information about my particular build.

• Genetics: Male, 5'10, 165 lbs (prev 200 lbs 🥶)
• Hobbies: Sleeping, Zone 5 Cardio, Softball, Mountain Hiking