AccelOpt

We are building self-improving agentic systems that ship silicon to intelligence.

Contact: Genghan Zhang (zgh23@stanford.edu)

Papers

• AccelOpt: A Self-Improving LLM Agentic System for AI Accelerator Kernel Optimization
• Adaptive Self-improvement LLM Agentic System for ML Library Development ICML 2025

Blogs

• AccelOpt+Fixer for Triton on H100
• AccelOpt: Democratizing AI Accelerator Programming with Self-Improving LLM Agents
• Adaptive Self-improvement LLM Agentic System for ML Library Development