{"data":{"posts":{"edges":[{"node":{"frontmatter":{"title":"Making One GPU Go Brrr: Reinforcement Learning in a Single CUDA Kernel","date":"Nov 2026","slug":"/megarl/","external":true,"category":"research"}}},{"node":{"frontmatter":{"title":"Compute-Aware Hybrid Attention Architecture Search","date":"Dec 2025","slug":"/blog/compute_aware_arch_search","external":null,"category":"research"}}},{"node":{"frontmatter":{"title":"Fast Humanoid Control with Flow Matching","date":"Dec 2025","slug":"/blog/flow_matching_robo/","external":true,"category":"research"}}},{"node":{"frontmatter":{"title":"Why PPO Beat GRPO at Teaching a Vision-Language Model to Solve Mazes","date":"Dec 2025","slug":"/blog/grpovsppo/","external":true,"category":"research"}}},{"node":{"frontmatter":{"title":"Parallelizing CDF","date":"Sep 2025","slug":"/blog/hack-project","external":null,"category":"research"}}}]}}}