About the job
The Seed Infrastructures team oversees the distributed training, reinforcement learning framework, high-performance inference, and heterogeneous hardware compilation technologies for AI foundation models.
Responsibilities
Design and build end-to-end reinforcement learning (RL) systems for large-scale models, covering rollout, training, evaluation, and deployment pipelines.
Develop scalable and fault-tolerant RL infrastructure that operates efficiently under dynamic workloads and heterogeneous compute environments.
Optimize distributed training performance across GPU clusters, improving throughput, resource utilization, and system stability.
Collaborate with cross-team researchers on targeted system–algorithm co-design to translate research ideas into robust, production-grade implementations.
Build tooling, monitoring, and debugging frameworks to ensure reliability and observability of large-scale RL training systems.
Qualifications
Minimum
Strong background in distributed systems, large-scale ML systems, or deep learning infrastructure
Experience building or optimizing large-scale training systems (e.g., RL, LLM, multimodal models)
Solid engineering skills in Python/C++ and familiarity with modern ML stacks (PyTorch, distributed training frameworks, etc.)
Experience with GPU optimization, parallelism strategies, and system-level performance tuning
Understanding of reinforcement learning workflows (rollout, policy update, evaluation loops)
Preferred
Experience with large-scale agent systems
Familiarity with system design under heterogeneous or dynamic workloads
Exposure to RL + LLM training or post-training pipelines