# Arjun Sahlot

I love computers. I’m studying them at [UW–Madison](https://arjunsahlot.com/notes). Last summer I worked on [Locus](https://trylocus.space) and [Summary Bot](https://trysummarybot.com), along with a bunch of fun AI experiments in the [lab](https://arjunsahlot.com/lab), while holding an ML engineering position at [Magic Hour](https://magichour.ai). The previous summer, a software engineering position at [MathWorks](https://mathworks.com).

Most of what’s here started as a question I wanted a real answer to. Some of it became finished work. The rest is still being figured out.

## Selected work

### Diffusion Maze Solver (2026)

Pathfinding by denoising instead of search. A U-Net trained from scratch takes a 23×23 maze and a field of noise, then walks a thousand reverse diffusion steps until the noise resolves into a path from start to goal. It runs live in the browser.

Stack: Python, PyTorch, Gradio

- Demo: https://maze-diffusion.arjunsahlot.com
- HF Space: https://huggingface.co/spaces/ArjunSahlot/diffusion-maze-solver
- Source: https://github.com/ArjunSahlot/diffusion-maze-solver

### Locus (2026)

A mobile workspace for working with AI. Each space pairs a conversation with a document the agent can edit directly, so the result of the work outlives the chat that produced it.

Stack: React Native, Expo, Next.js, AI SDK

- trylocus.space: https://trylocus.space

### ChessBench (2026)

A standardized harness that asks language models to write chess engines, then plays the engines against each other. Roughly three thousand games of round-robin turn a subjective question into an Elo number.

Stack: Python, C++, Next.js, SQLite

- chessbench.live: https://chessbench.live
- Source: https://github.com/ArjunSahlot/ChessBench

[All projects](https://arjunsahlot.com/projects)

## Lab

A few experiments going, none of them ready to be looked at yet.

- GPT-2, but better
- notes/tools collection

[All experiments](https://arjunsahlot.com/lab)

## Experience

- Magic Hour (YC W24) — ML engineering, 2026
- MathWorks — Software engineering, 2025
- UW–Madison — B.S. Computer Science, 2024–present

## Writing

- [Playing around with FrameFT](https://arjunsahlot.com/writing/playing-around-with-frameft) — Experiments, Oct 15, 2026: FrameFT fine-tunes a language model by training just 24,000 numbers through a carefully built matrix. I swapped that matrix for a random one to see what's actually doing the work.
- [The gap in human-agent communication](https://arjunsahlot.com/writing/human-agent-communication) — Thoughts, Oct 2, 2026: Models learn to write from text meant for strangers. What you want is a collaborator who already knows the context.

[All writing](https://arjunsahlot.com/writing)

## Elsewhere

Email is the best way to reach me — iarjun.sahlot@gmail.com. I’m also on [GitHub](https://github.com/ArjunSahlot), [X](https://x.com/arjunsahlot), and [LinkedIn](https://linkedin.com/in/ArjunSahlot).