Machine Learning Engineer, Core Experimentation

About the Team The Statsig team within OpenAI builds the experimentation, feature rollout, dynamic configuration, and analytics systems that help OpenAI ship products with speed, safety, and evidence. Our work sits on the critical path for how product, engineering, research, and go-to-market teams learn from real-world usage and make high-confidence decisions. Statsig began as an independent company focused on helping builders move faster through trustworthy experimentation and feature management. After Statsig joined OpenAI, the team began the next chapter: bringing that deep product expertise, customer intuition, and mature platform infrastructure into OpenAI as the experimentation and rollout platform for every product we ship. Today, we support teams across ChatGPT, Codex, model measurement, consumer experiences, business subscriptions, developer products, and the shared infrastructure that connects them. These teams rely on Statsig to safely introduce new capabilities, compare product and model behavior, measure impact, and roll changes forward or back with confidence. We are at a defining moment in the platform journey. OpenAI has the data, product surface area, and pace of innovation to learn faster than almost any organization in the world, but that potential only becomes real if teams can experiment responsibly, measure clearly, and roll out changes safely. We are evolving experimentation systems to help teams learn from product behavior and make better evidence-based decisions. Based out of OpenAI’s Bellevue office, we are a close-knit team that values in-person collaboration, urgency, craft, and impact. We build for other builders, and the best version of this team is one where every OpenAI product team can move faster because the experimentation and rollout layer is dependable, fast, and easy to use. About the Role We are looking for a Machine Learning Engineer to lead the technical direction for ML-powered experimentation and insights capabilities. You will build production systems that learn from privacy-protected product and experimentation data to generate evidence-backed insights and support decision-making, and help teams decide which ideas are worth testing live. This is an end-to-end, 0-to-1 role. You will work across ML modeling, retrieval and LLM systems, statistical methods, simulation, data and training pipelines, backend services, and user- and agent-facing product experiences. The hard part is not merely producing a plausible answer. It is making each insight and prediction traceable, calibrated, useful, and safe enough to influence real product decisions. Live experiments remain the source of causal validation. You will design systems that make uncertainty explicit, backtest against historical outcomes, compare predictions with online results, learn from misses, and abstain when the evidence is weak. You will preserve clear review, permission, and approval boundaries as automation becomes more powerful. You will collaborate closely with teams building ChatGPT, Codex, model measurement workflows, consumer products, Growth, business subscription experiences, developer products, and shared infrastructure. You will turn their most important learning and decision problems into general platform capabilities that can support the full company. In This Role, You Will Set and execute the technical roadmap for Generative Insights and Predictive Experimentation, from early prototypes through production adoption. Build cross-experiment learning systems that retrieve and synthesize historical experiments, detect recurring effects and segment behavior, reanalyze prior results when data or methods improve, and generate hypotheses with clear evidence and provenance. Develop predictive models and simulation workflows, including simulation-based evaluation approaches, to estimate likely impact, affected segments, regression risk, and uncertainty before a full live experiment. Create high-quality datasets and feature or retrieval pipelines from exposures, events, metrics, experiment metadata, and replay data, with strong lineage, freshness, privacy, and data-quality controls. Establish rigorous evaluation through offline benchmarks, backtests, calibration, drift monitoring, prediction-to-outcome comparisons, and explicit failure or abstention behavior. Turn models into durable product, API, and agent workflows that move from an insight to experiment design, approval-gated action, and measured learning. Partner deeply with data science and product teams on experiment design, causal inference, sequential decision-making, variance reduction, and the boundary between prediction and causal evidence. Build reliable services and intuitive workflows so sophisticated ML capabilities are understandable and useful to teams making high-stakes product decisions. Provide technical leadership across engineering, product, data science, and research partners, and raise the bar for production ML quality across the pla