Every great recommendation on Prime Video starts with data. When millions of customers open the app to decide what to watch next, the quality of what they see depends on how fresh, complete, and reliable the data behind the scenes is. PV Personalization & Discovery helps customers find new titles and rediscover the content they love, delivering personalized, relevant recommendations tailored to each customer's context, schedule, and environment. And behind every one of those recommendations is the data platform that makes them possible.
That's where you come in. As an SDE2 on PVPD's Data Platform, you'll build and own the online systems that turn raw customer behavior into fresh, reliable features and training datasets for our personalization and ranking models. You'll work on the near-real-time signal path that takes a customer's latest actions (a stream, a click, a search) and makes them available to models in seconds instead of hours, directly shaping what customers see on the storefront while their intent is still fresh.
This is a systems-heavy role at the intersection of data and scale. You'll design and operate low-latency feature-serving systems, streaming ingestion and processing pipelines, and the unified, extensible datasets that let a single investment in data quality, coverage, and freshness lift every downstream recommendation model, experiment, and science evaluation at once. You'll own the data contracts that keep online serving and offline training consistent, and the production reliability of these systems (freshness SLAs, availability, monitoring, and graceful degradation when upstream signals lag) because once a feature or dataset becomes a model dependency, it has to be trustworthy every time.
You'll collaborate with a talented team of engineers and scientists across PV Personalization & Discovery: partnering with applied scientists to understand which signals move the needle, and with modeling teams to deliver those signals and datasets at incredible scale and speed. If you're excited about building the real-time data foundations and enabler datasets that power AI-first personalization for one of the largest streaming audiences in the world, we'd love to talk.
Key job responsibilities
- Build and operate the online feature-serving systems that deliver customer-behavior signals to Prime Video's personalization and ranking models within a strict latency budget.
- Design and scale near-real-time streaming pipelines that take a customer's latest actions (e.g. streams, clicks, searches) from event to model-ready feature in seconds, meeting freshness SLAs end to end.
- Build the unified, extensible datasets and enabler data products that let a single investment in coverage, quality, and freshness lift every downstream model, experiment, and science evaluation at once.
- Own the data contracts and online/offline consistency that keep training and serving in sync, so features behave the same in production as they do in offline evaluation.
- Own production reliability for these systems including monitoring, alerting, freshness and availability SLAs, and graceful degradation when upstream signals lag.
- Take work from an ambiguous problem statement through design, launch, and measured impact on model and customer outcomes.
- Partner directly with applied scientists and modeling teams to turn the signals that move the needle into reliable, production-grade data.
- Work in an agile environment to deliver high-quality software.
About the team
In Prime Video Personalization and Discovery, our mission is to show a customer the right video content, in the right place, at the right time. We tailor PV for a diverse and global audience, so customers world-wide with different tastes and backgrounds can find something to watch and enjoy on every visit. We delight Prime subscribers by leaning in to the Amazon originals and exclusives that shape our brand and differentiate us. We grow our relationships with customers by leveraging our deep understanding of them to provide relevant and timely recommendations.
