Software Development Engineer, Demand Forecasting & Guidance

Live sports and Streaming TV are the fastest-growing premium inventory in advertising, and they are the hardest to forecast. A regular-season game and a playoff elimination game are the same length and the same ad load, but they are not the same business. Amazon DSP's Live Events Optimizer (LEO) launched in December 2025 and is scaling from its first cohort of advertisers to hundreds; every one of them asks the same two questions before they commit budget - *how much inventory will there be, and when will it show up?* Answering those questions accurately is what our team builds. We are looking for a Software Development Engineer to build and operate the systems that produce and serve those forecasts across Live Events and Streaming TV. This is a hands-on engineering role at the intersection of large-scale data pipelines, real-time services, and applied ML. You will write the code that ingests broadcast schedules and live game state from multiple third-party sources, resolves them into a canonical view of what is airing and when, and turns forecasted supply into signals that pacing and bidding act on. You will work directly with applied scientists and data scientists who are pushing on the modeling side (e.g. tabular foundation models with LLM-derived semantic priors, contention and win-rate modeling, zero-shot forecasting for inventory with no history) and your job is to make that science real: productionized, monitored, cost-aware, and fast enough to matter during a live broadcast. Several of these systems are on the critical path for live inventory, so you will also raise the operational bar: telemetry, accuracy dashboards and guardrails that fail safely. Key job responsibilities - Design, build, test, and operate components and services within our forecasting and data platform, owning them from design through production. - Write high-quality, maintainable, well-tested code, and participate actively in design and code review. - Break down assigned projects into workable tasks, deliver them on a predictable schedule, and communicate progress and risks clearly. - Partner with applied and data scientists to productionize models - building and maintaining training, inference, and monitoring pipelines, and helping translate offline experiments into running systems. - Build the instrumentation and dashboards that show whether our forecasts are accurate, timely, and cost-efficient, and investigate when they are not. - Participate in the team's on-call rotation and contribute to operational excellence - runbooks, alarms, and reducing recurring toil. - Learn the advertising and live-sports domain deeply enough to make good independent judgment calls, and share what you learn with the team.