GRAND ARENA
AI Training Infrastructure
My Contributions
I designed and built reinforcement learning infrastructure that reduced the AI model iteration cycle by 60 days.
I led the design and implementation of systems that managed parallel Unity simulations for training and evaluating AI agents, including dynamic environment configuration, curriculum progression, observations, and reward systems.
I provisioned Linux virtual-machine infrastructure on Google Compute Engine for simulation and debugging, and used Python, C#, JSON, and YAML to make agent-training environments configurable at runtime.