Most of what I build in my own time runs on your machine, with your keys and your data.
An open-source AI coworker that lives in your Slack and runs on your own servers. Mention it in a thread and it plans, works across your tools, writes and runs code in a sandbox, remembers how your team works, and shows every step and every cent.
Bring your own model, your own infra, your own tools. Free forever.Kortny on GitHub
You star repos and forget them. StarDash remembers, and tells you which new ones your taste says to look at.
StarDash on GitHubA private practice space built around your own work. Your AI writes the problems, reviews what you build, and interviews you on it.
Whetstone on GitHubOver 2,100 US stocks scored on technical and fundamental factors, so the interesting ones float up.
deepdiver on GitHubI build, optimize and run Longboard's trading systems, mostly on my own. That covers the platform that turns market data into each day's trades, the pipelines that keep our books in step with custodians, and the cloud under all of it.
Lately it also means putting language models to work on real operations, with guardrails, so the team can ask the book a question in plain English and get an answer they can trust.
Trend followers are good at stepping aside when markets fall apart. Getting back in is the hard part. By the time individual stocks make fresh highs again, much of the rebound has already happened.
I'm studying what a rules-based re-entry should look like for a long-only stock portfolio, a case the published research hasn't covered. That means rebuilding past crashes from price history, testing simple rules against them, and writing it up as a paper.
A cooking and deckbuilding roguelike set in the Mahabharata's hidden year, when Bhima disguises himself as Ballava, cook to the court of King Virata. I write the systems and the story. My partner does the art, the music, and the recipes.
It will be completely free to play.
It's drawn like the mythology comics many of us grew up reading in India: bold ink, flat colour, the same look by day and night, lit differently.
I have been playing video games for about 25 years, since one had to type P‑R‑I‑N‑C‑E in the command line to start the original Prince of Persia.June 2022
Reward loops, battle royales, and a lifelong gamer watching his own frustration.
Addictive feeds on one side, dark patterns on the other, and the user pulled between them.
Social proof, scarcity and the other biases behind the cart, and where a nudge should stop.
Three apps for three very different users, and why you build for theirs, not yours.
Authority, reward, guilt and the urge to help: the levers a con artist pulls.
Fewer files, fewer dependencies, a shorter diff. Simple code reads as intent.
Nothing ships until someone has watched it work. In AI systems, checking the answer beats fetching more context.
Legibility beats elegance, in a game mechanic or an admin page.
A prompt that is patched every week needs a test suite, not another patch.
Self-hosted, bring your own model, and no company that can take it away from you.
Systems can have beautiful names. The button still says what it does.
