MS AI @ Northeastern (Silicon Valley) | GPU Kernel & Inference Optimization (CUDA / CuTe DSL) | Silicon Valley Fellow W25
📝 I write articles on FreeCodeCamp · 📫 Reach me at [email protected]
I make AI models run faster and cheaper by writing GPU kernels close to the metal.
Currently working on GPU kernel optimization with CuTe DSL.
What I'm building
- ⚡
fast_trimul- drop-in Fused Triangle Multiplicative Update kernel for AlphaFold3-family models (OpenFold-3, Boltz, Chai, Protenix). Up to 4.5–6.8x faster inference, 2.2–2.4x less peak VRAM. - 🧠 CUDA / CuTe DSL kernels - softmax attention 62x faster (22.8ms → 0.36ms) via memory-hierarchy tuning; #2 on LeetGPU's leaderboard.
- 🔬 Research in tabular machine learning and synthetic data generation (multiple articles under review / in final writing).
- 📚 Author of The Math Behind AI (160+ ⭐) and How to Build Optimal AI Agents - 250k+ readers across my writing.
Foundation: Hardware-level + deep math. In my ECE bachelor's I built a CPU from scratch (transistor up), with a strong math base in linear systems, calculus, differential equations, complex analysis, PDEs, and functional analysis.
