Independent AI Researcher & Builder
I’m an independent researcher who believes that the most valuable contributions in AI come from building principled theoretical frameworks — ones that don’t just explain existing phenomena, but actively guide the design of new methods. My work sits at the intersection of theory and methodology: I care as much about why something works as how to make it work better.
The same philosophy carries into how I build. I look for the gaps that others treat as given, and build from there.
Research #
Window is Everything (2025) 17K+ views · 4K+ downloads · Hacker News Front Page (Sep 2025)
A unified theory of neural operations that decomposes any operation into three orthogonal components: Path (P), Shape (S), and Weight (W).
Projects #
** Agenlus** — A community platform for reinforcement learning. Browser-based RL training via WebGPU + Pyodide, no install required.
** FactorizedAttention** — PyTorch implementation of Factorized Attention mechanisms.
Get in Touch #
- Email:
[dafaafafaf33@gmail.com](mailto:dafaafafaf33@gmail.com) - Twitter / X:
[@salam341353](https://x.com/salam341353) - GitHub:
[Kim-Ai-gpu](https://github.com/Kim-Ai-gpu)