About the job
Amazon Search is building a first-of-its-kind AI-powered visual search experience that lets customers describe products they're imagining, instantly see AI-generated images in response, and tap those images to search for matching products to shop. We are transforming the search engine into a shopping engine by leveraging advances in generative AI and multimodal understanding.
Responsibilities
Design, train, and optimize generative AI models for real-time product image generation, ensuring outputs meet strict latency requirements while maintaining high visual quality and query alignment.
Develop multimodal retrieval systems that connect AI-generated images to Amazon's billions-scale product catalog, optimizing for recall and ranking relevance across product categories.
Build LLM-based classifiers for visual intent detection, query understanding, and safety filtering within real-time latency budgets.
Advance AI safety science through defense-in-depth approaches including embedding-space classifiers, adversarial data engines, and post-generation content moderation.
Design and execute large-scale online experiments to measure impact on customer engagement, search success, and business metrics, defining evaluation frameworks that combine automated metrics with human judgment.
Collaborate with engineering, product, and design teams to architect GPU-intensive inference pipelines serving real-time traffic at scale, and contribute to Amazon's scientific community through publications and patents.
Qualifications
Minimum
PhD, or Master's degree and 4+ years of CS, CE, ML or related field experience
Experience programming in Java, C++, Python or related language
Experience in state-of-the-art deep learning models architecture design and deep learning training and optimization and model pruning
Preferred
Experience using Unix/Linux
Experience in professional software development
Experience developing and implementing deep learning algorithms, particularly with respect to computer vision algorithms
Have publications on top-tier conferences, such as CVPR, ICCV, ECCV or NeurIPS