│ -->

     

◇  research I build and align multimodal foundation models — teaching systems to perceive across vision, language, and audio, then sharpening their behavior through post-training and reinforcement learning, and serving them fast. VLM — visual perception & grounded reasoning Post-training — SFT, preference alignment, distillation RL — RLHF / RLVR, reward modeling, agentic policies KV cache — efficient long-context inference & serving TTS — expressive, low-latency speech synthesis 3D — spatial reasoning & generation

◆  currently class Researcher: name = "clos" role = "ai researcher" focus = ["vlm", "post-training", "rl"] systems = ["kv-cache", "tts", "3d"] def philosophy(self): return "less noise, more signal" Chasing the smallest model that still understands.

◈  toolbox

  

_{— less noise, more signal —}