building autonomous systems at Vesper Dynamics. RL, edge deployment, alignment research.

San Jose ↔ Tokyo · colonel1223.net

most of my time right now goes to:

• real-time ML inference on edge hardware (quantization, pruning, latency optimization)
• hierarchical RL policies for embodied agents
• figuring out why alignment guarantees break at scale (formal models)

some things I've built:

• learned-reranker — hybrid retrieval + neural re-ranking, +36% NDCG@10
• conformal-multimodal — distribution-free uncertainty quantification
• CHIMERA — 847K traces showing hallucination is information-theoretic
• agentic-rag-diagnostics — closed-loop retrieval agent

python c++ pytorch rl edge deployment

日本語も話せます