Agent Infra

• PenguinHarness  ⭐ 2.1k
  • Role: Chief Architect.
  • A local-first platform where agents autonomously build and improve other agents across the full app lifecycle, at far lower cost.
• GDPevo  ⭐ 62
  • Role: Project Lead.
  • The first benchmark for agent self-evolution on economically valuable (GDP-related) tasks, lifting test accuracy by 17–22% while cutting token cost.
• Awesome-RSI  ⭐ 55
  • Role: Project Lead.
  • A curated hub of benchmarks and methods for Recursive Self-Improvement (RSI)—agents that improve themselves.

LLM Infra

• RedKnot  ⭐ 1.7k
  • Role: Project Lead.
  • A long-context LLM inference engine on SGLang that cuts prefill FLOPs by 50%–70% for 1.35x–2.2x faster TTFT at near-lossless accuracy.