Technical Program Manager, Research Compute Services
OpenAI · San Francisco
About this role
**ABOUT THE TEAM** Research Compute Services builds the systems that help researchers use OpenAI’s compute fleet effectively. We partner with Research and Compute Infrastructure to support workloads across the compute stack, including pretraining, synthetic data generation, sampling, research inference, and reinforcement learning. Our work connects changing research needs with capacity planning, allocation, and scheduling. We build tools and services that help researchers express their compute needs, access the right resources, and run their workloads effectively. As the fleet and our research ambitions grow, we continually improve how compute is planned, allocated, and used. **ABOUT THE ROLE** You’ll be part of an engineer-first TPM team, partnering closely with Mahmoud and the Research Compute Services team. Your primary responsibility is to build and own the tools, data, and platform that turn Research plans and workload signals into concrete compute demand and manage allocations across the fleet. This is a hands-on technical role. You’ll build alongside researchers and engineers—working directly with code and data to prototype solutions, automate workflows, and deliver tools that teams use every day. You’ll combine technical judgment and program leadership to connect capacity planning with deployment, scheduling, and optimization systems that make allocated compute usable. You’ll work across a broad range of Research workloads, understanding their different requirements and ensuring those needs shape the platform. **IN THIS ROLE, YOU WILL:** - **Build with Research:** Work directly with researchers across pretraining, research inference, and other workloads; translate their needs into improvements across compute services, tooling, and scheduling. - **Own Research demand and allocation systems:** Own the roadmap and delivery of tools, data, and platform used to capture demand, support allocation decisions, and manage allocations—from requirements and design through adoption and ongoing improvement. - **Turn research plans into actionable demand:** Develop systems that translate evolving plans and workload requirements into concrete capacity needs (resource requirements, timing, and constraints). - **Connect capacity supply and demand:** Partner with Capacity Planning, Research, and infrastructure teams to reconcile demand with available supply; build tools/workflows to evaluate scenarios, manage changes, and make allocation tradeoffs visible. - **Build reliable data and useful tools:** Create data models, views, and automation that show what researchers need, what capacity is available, what has been allocated, and how it’s being used; improve self-service workflows for requesting and managing compute. - **Connect allocations to workload execution:** Partner with engineers working on Kubernetes deployments and scheduling systems so allocation plans reflect workload constraints and translate into resources researchers can use. - **Improve allocation and scheduling efficiency:** Evaluate optimization approaches (fleet and local scheduling); define constraints/objectives, assess tradeoffs, and measure improvements in fleet utilization. - **Find and address systemic gaps:** Use workload and fleet data to identify bottlenecks, unused capacity, and recurring manual work; build durable improvements. - **Lead technical delivery across teams:** Own plans, dependencies, and decisions across Research, Research Compute Services, Capacity Pl
Listing freshness
CronJobs last confirmed this listing 1h ago. If its source stops confirming the opening for seven days, this page is removed from active inventory.