About the RoleWe are seeking a Technical Program Manager to build the flywheel that helps ChatGPT multimodal products learn from real-world usage and improve quickly. You will lead programs spanning production-signal mining, evaluation and data pipelines, research-to-production parity, multimodal capacity planning, and complex cross-functional dependencies for voice and image-generation launches.
You will work closely with product engineering, research, Human Data, inference and capacity teams, safety partners, and external vendors or product partners. Success requires technical depth, strong systems thinking, comfort with ambiguity, and the ability to turn fragmented or manual work into durable mechanisms that teams adopt.
This role is based in San Francisco, CA. We use a hybrid work model of 3 days in the office per week and offer relocation assistance to new employees.
In this role, you will:- Build a system for mining production conversations and product signals to identify representative multimodal workflows, user needs, and failure modes.
- Establish and maintain evaluations for the highest-priority multimodal behaviors and use cases, with clear coverage, quality standards, and ownership.
- Package production signals into decision-ready data and evaluations that research teams can use to improve model behavior.
- Measure whether model, prompt, configuration, and product changes produce meaningful improvements in multimodal evaluations and user outcomes.
- Close gaps between research and production environments, including system prompts, sampling behavior, multimodal configurations, inference differences, and other sources of parity drift.
- Create a repeatable process for reproducing product failures with research partners and validating fixes in the shipped experience.
- Lead multimodal capacity planning by forecasting demand, translating it into GPU and serving needs, and managing headroom and reallocation tradeoffs for voice and image-generation workloads.
- Improve the tooling and operating processes used to plan, launch, and operate multimodal capabilities as demand and model behavior evolve.
- Coordinate targeted multilingual data collection across research, Human Data, and external vendors.
- Drive cross-functional programs that multimodal launches depend on, including multimodal actor recruitment and selection and voice-related product partnerships across vehicles, smart speakers, and headphone ecosystems.
- Create clear operating cadences, decision rights, metrics, risk management, and executive-ready communication across complex, time-sensitive programs.
You might thrive in this role if you:- Have led technically complex programs across machine learning, multimodal products, model evaluation, data pipelines, inference, capacity, or large-scale product infrastructure.
- Can move fluently between user-facing product behavior and the underlying model, evaluation, configuration, serving, and capacity systems.
- Have built mechanisms that convert production signals into prioritized evaluations, data, engineering work, and measurable product improvements.
- Can reason credibly about GPU demand, serving constraints, latency, reliability, quality, and launch tradeoffs.
- Have closed research-to-production gaps and can drive structured debugging and validation across teams with different environments and incentives.
- Turn manual or fragmented workflows into scalable tooling, clear ownership, and durable operating practices.
- Lead effectively across research, engineering, product, operations, Human Data, and external partners without relying on direct authority.
- Communicate crisply, make ambiguity tractable, and use metrics and evidence to drive decisions.
- Thrive in ambiguous, scaling environments and can bring order to complex cross-functional work without losing pace.
- Care about OpenAI's mission and about expanding responsible access to advanced AI systems.
