Design and build end-to-end reinforcement learning training infrastructure: rollout and reward pipelines, actor-learner architectures, distributed orchestration, monitoring, and GPU utilization improvements, partnering with researchers to productionize RL training for multimodal models.

Design, optimize, and scale low-latency, high-throughput inference serving systems for large multimodal models. Implement quantization, batching, speculative decoding, KV cache management, multi-GPU/multi-node parallelism, autoscaling, and load balancing while optimizing GPU utilization, reliability, and observability for production deployments and research evaluation.

Build and own the full-cycle technical recruiting function: source and close top AI/ML and systems talent, design hiring processes and ATS workflows, advise founders on role design and compensation, shape candidate experience and employer branding, and scale the company’s talent pipeline from the ground up.