Jupyter-like Notebooks for Shaders
web based GLSL shader notebooks: write fragment shaders in editable cells with live previews — like a Jupyter notebook but for shaders.
Armeet Singh Jatyani
Hi, I'm Armeet! I'm a co-founder at Arcten and an ex-Caltech AI researcher (vision models) that published in the top journals (CVPR, NeurIPS). My website ...
Latest posts
The Game Balancing Mechanic
When you're building a startup, you have this irrational belief that you will win. Despite loaded incumbents with oceans of capital and capable competitors, you can't shake the feeling that winning is entirely within reach. It takes an irrational amount of hubris to look at the world around you and believe you can do things better. If you're not careful, it can turn into arrogance. But the…
Sandbox Coding Agents Suck (right now)
Background sandbox coding agents like Cursor, Codex, and Copilot have not really taken off. As a YC founder who uses coding agents daily, here is why. 1) They cannot run the code or tests If I see one more PR from Copilot that does not compile or fails a basic typecheck, I am going to lose it. What is missing: Coding agents should run on machines with the full developer environment. That means…
The Faster Shell
tldr: I switched from fish to plain old bash because it's fast and does what I need it to do 2 years ago, me and my roommate decided to learn vim. We built our very own neovim configs, learned about LSPs, and built a compiler in OCaml. It felt amazing and was also one of my favorite memories from Caltech. This led me down a certain road. I spent a lot of time trying the latest cutting edge tools:…
Don't Become the Machine
I was recently recommended a YouTube video with the following title: Only a slave quantifies its existence through productivity '''''',,,,,,,,,,,'''''.''''''........................................................................... ......................................... ... '''''',,,,,,,,,,'''''''''''''''...........................................................................…
Web MRI Volume Renderer in Rust
Over the weekend, I made a GPU-accelerated 3D MRI volume renderer written in Rust that compiles to WASM and runs directly in web. Here’s my repo: github . I’m open to PR’s! You should go checkout the demo application website and video (15s) . Back when I was a researcher working on AI models for 3D medical imaging, one of the problems I faced frequently was visualizing extremely dense (512³) MRI…
Importance of Writing
Writing is becoming increasingly automated and the web is littered with AI generated content. While I'm not opposed to reading a valuable blog post on some engineering topic, even if it's been written by AI, the truth is that most of this content (currently) is crap. Using AI to assist your writing and help sculpt out a piece of work is fine, but most people don't use it this way. The internet is…
The Reasonable Effectiveness of the Web
tldr: The web gets you 90% of the way there, at 10x the speed. You don't have to download an application, assets, frameworks, or plugins. Everything is plug and play. You load a site in Sure, websites may not be the end-all performance platform, and there are apps that strictly need to build on the desktop. But the small percentage of these applications is getting even smaller with web assembly…
Telegram Alerts for Training
I often use Weights & Biases for training. Others might use something like Tensorboard or mlflow . While these are great for tracking experiments and sweeps or monitoring logs, sometimes you want a simple alert system that'll wake you up when a run starts or fails (but with a little more control so you can quickly get a view on experiment status). While SLURM allows you to setup email alerts,…
Multi-Node Training on TACC Vista
Introduction SLURM Script Environment Variables Testing test_mpi.py test_simple_distributed.py test_lightning_mpi.py Multi-node training can be tricky. When training on multiple nodes, machines must communicate with one another, synchronizing gradients and state. Most frameworks handle most of the heavy lifting behind distributed training algorithms (like torch's DDP implementation). However,…