LLM systems ยท embodied AI

YU AORI

Focused on training/inference efficiency, embodied intelligence, and dependable distributed tooling.

Portrait of YU AORI

Core tech stack

LLM Training & Optimization

  • PyTorch FX / Hooks
  • Custom CUDA kernels
  • DeepSpeed / ZeRO
  • vLLM & TensorRT-LLM
  • LoRA / Q-LoRA pipelines

Systems & Infrastructure

  • Linux
  • Docker & Kubernetes
  • Redis / ElasticSearch
  • Observability (Prometheus / Grafana)

Full-Stack & Tooling

  • TypeScript / Node.js
  • React / Next.js
  • Monorepo build systems
  • RxJS / Mutative state
  • tRPC + GraphQL Codegen
  • Secure QuickJS sandboxes

Embodied AI & Simulation

  • MuJoCo / Genesis
  • Blender synthetic data
  • GRPO reinforcement learning
  • World model research
  • Multimodal VQ encoders