Amazon's PRISM team is seeking an innovative Applied Scientist to build the intelligence layer powering the Catalog Diagnostic Assistant — a conversational AI agent that unifies Amazon's fragmented catalog diagnostic experience into a single natural-language interface. Rather than switching between legacy tools, internal users ask CDA a question in plain language and CDA coordinates across data sources, reasons through complex diagnostic workflows, and returns a combined answer with source citations.
This role sits at the intersection of Generative AI, agentic architectures, and large-scale information retrieval applied to the world's largest product catalog. You will design and build the scientific core of an agent that autonomously investigates catalog anomalies — diagnosing why products aren't live, why attributes aren't publishing, or why matching decisions went wrong — across billions of products, petabytes of multimodal data, and dozens of marketplaces.
You will be the founding scientist for the CDA product, defining the research agenda for agentic diagnostics, developing novel approaches to skill-based reasoning and tool orchestration, and owning the full lifecycle from problem formulation through production deployment at Amazon scale. You will pioneer advanced GenAI solutions that power next-generation agentic experiences, working in a collaborative environment where you can experiment with massive data from the world's largest product catalog and tackle problems at the frontier of AI research.
Key job responsibilities
- Formulate open research problems at the intersection of GenAI, agentic reasoning, and large-scale catalog diagnostics — defining how an autonomous agent should decompose, investigate, and explain complex catalog issues
- Design and develop novel agentic architectures (skill planning, tool selection, multi-step reasoning, chain-of-thought verification) that enable CDA to autonomously resolve diagnostic workflows that traditionally required manual expert investigation
- Build and optimize retrieval-augmented generation (RAG) systems over Amazon's catalog data, ensuring the agent retrieves the right evidence from the right data sources to ground its diagnostic answers
- Advance the science of efficient model deployment — developing distillation, compression, and LLM serving optimization strategies that preserve diagnostic reasoning quality in production-grade architectures while reducing latency and cost
- Make frontier models reliable for autonomous decisions — advancing uncertainty calibration, confidence estimation, and interpretability methods so CDA's agentic diagnoses can be trusted at scale
- Own the full research lifecycle from problem formulation through production deployment — designing rigorous experiments, iterating rapidly, and seeing your research directly improve diagnostic accuracy and coverage
- Partner closely with CDA engineers to translate research prototypes into production systems serving thousands of daily diagnostic sessions
- Shape the team's research vision by defining technical roadmaps that balance foundational scientific inquiry with measurable product impact
- Mentor engineers on advanced ML/GenAI techniques, experimental design, and scientific rigor
About the team
We are a GenAI science team within Amazon's Selection and Catalog Systems (ASCS) organization. Our mission is to advance state-of-the-art Generative AI to deeply understand, uniquely identify, and intelligently diagnose every product at Amazon scale. We push the boundaries of multimodal LLMs, agentic systems, and generative AI to solve foundational catalog challenges — from product identity and semantic matching to automated catalog diagnostics and intelligent catalog analytics.
Our scientists work across multiple high-impact problem spaces: establishing canonical product identity across billions of items, powering next-generation agentic experiences for catalog operations, and building AI systems that make Amazon's catalog self-diagnosing and self-healing. We operate at the frontier of AI research applied at unprecedented scale — petabytes of multimodal data, millions of sellers, dozens of languages, and infinite product diversity.
This is a collaborative, fast-moving environment where you can experiment with massive data from the world's largest product catalog, tackle problems at the frontier of AI research, rapidly implement and deploy your ideas at scale, and publish your findings at top venues.
