Ilyaas Kapadia

I work on reinforcement learning systems — fast C and CUDA environments where agents learn to play games. Current obsessions: a speedrun policy for all 11 levels of the 2001 Nokia game Bounce, and squeezing every step per second out of a single RTX 5070.

I interned at Kernel (YC S25), working on browser infrastructure for AI agents.

Before this, I was a mountain biker, a salesperson, and a businessperson. I write about what I'm learning.

Projects

• Countdown — try a 0.8B model trained with reinforcement learning on arithmetic puzzles, running in your browser
• Slither
• Bounce — Nokia’s 2001 classic, rebuilt for the browser and played by a trained policy
• kstack — E-ink-first Android client for driving a fleet of headless omp agents over Tailscale SSH

Internship @Kernel (YC S25)

• Cut the server CI workflow from 17m33s to 7m29s — PR #299
• Slimmed the Chromium runtime images by up to 366 MB; headful cold build 347s to 268s — PR #274
• terraform-provider-kernel — Terraform provider for Kernel’s browser infrastructure

Contact

I love meeting new people, and reply to every human-written email. Say hello.

Home | Writing | Internship guide | Resume | Email