Loading...
Applied Scientist II, Trustworthy Shopping Experience (TSE)
Amazon
On-sitestaffPosted May 12, 2026
About the Role
Design, build, and productionize deep agentic AI systems with multi-step reasoning and multimodal intelligence, leveraging advanced LLM techniques to automate investigation processes and enhance trust and safety for customers.
Requirements
Lead the development and productionization of next-gen agentic AI systems using Generative AI, LLMs, and multimodal intelligence to automate complex manual investigation processes for Amazon's Trustworthy Shopping Experience.
Full Job Description
Are you passionate about solving complex business problems at scale through Generative AI? Do you want to build intelligent systems that reason, act, and learn from minimal supervision? Are you excited about taking innovative AI solutions from proof-of-concept to production? If so, we have an exciting opportunity for you on Amazon's Trustworthy Shopping Experience (TSE) team.
At TSE, our vision is to guarantee customers a worry-free shopping experience by earning their trust that the products they buy are safe, authentic, and compliant with regulations and policy. We give customers confidence that Amazon stands behind every product and will make it right in the rare chance anything goes wrong. We do this in close partnership with our selling partners and empower them with best-in-class tools and expertise required to offer a high-quality selection of compliant products that customers trust.
As an Applied Scientist, you will lead the development of next Gen AI solutions to automate complex manual investigation processes at Amazon scale. You will work on some of the most fascinating challenges in applied AI—building systems that reason and act autonomously, learn rich representations from structured and relational data without extensive labels, adapt rapidly from limited examples, improve through feedback and interaction, seamlessly connect visual and textual understanding, and compress complex model capabilities into efficient, deployable systems. Your innovations will deliver significant impact to cost-of-serving customers while maintaining the highest standards of trust and safety.
This role offers end-to-end ownership—from initial research and proof-of-concept through production
deployment. You will see your innovations serving hundreds of millions of customers within months, not years.
Key job responsibilities
• Design and build expertise deep agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence with capabilities to handle feedback with long term as well as short term memory mechanisms.
• Design and build expertise agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence with capabilities to handle feedback with long term as well as short term memory mechanisms.
• Productionize large scale models built on top of SFT (Supervised Finetuning) and RFT (Reinforced fine tuning) approaches (GRPO with RLVR, Process/Outcome Reward Models), few shot approaches (Contrastive, Prototypical) based on multimodal datasets
• Enhance on existing Automatic prompt optimization techniques (GEPA & beyond) towards agentic optimization given the ground truth datasets to improve agentic planning.
• Build novel production ready Finetuned transformer architectures (using LORA/Q-LORA/LLM-JEPA etc) and conventional supervised & unsupervised ML solutions to aid the multiple potential automation requirements
• Identify customer and business problems at project level; invent or extend state-of-the-art approaches for complex LLM workflows involving unstructured text, documents, images, and relational data
• Author or co-author research papers for peer-reviewed venues; serve as PC member at conferences when aligned with business needs
• Prototype rapidly, iterate based on feedback, and deliver components at SDE 1+ level that integrate directly into production-scale systems
• Engineer efficient systems balancing model capability, deployment cost, and resource usage; write significant code demonstrating technical excellence and maintainability
• Scrutinize algorithm and software performance for improvements; resolve root causes leaving systems more maintainable
• Contribute to tactical and strategic planning—team goals, priorities, and roadmaps—while providing architectural guidance for AI systems
• Participate in engineering best practices with rigorous peer reviews; communicate design decisions clearly and participate in science reviews
• Train new teammates & interns on component construction and integration; mentor less experienced scientists and participate in hiring processes
About the team
Investigation technology Product team in TSE is responsible for the human-in-the-loop products and technology used in the risk investigations at Amazon. The team is also responsible for reducing the cost of performing the investigations, by automating wherever possible and optimizing the experience where manual interventions are needed. The team leverages state-of-the art technology and GenAI to deliver the products and associated goals.
At TSE, our vision is to guarantee customers a worry-free shopping experience by earning their trust that the products they buy are safe, authentic, and compliant with regulations and policy. We give customers confidence that Amazon stands behind every product and will make it right in the rare chance anything goes wrong. We do this in close partnership with our selling partners and empower them with best-in-class tools and expertise required to offer a high-quality selection of compliant products that customers trust.
As an Applied Scientist, you will lead the development of next Gen AI solutions to automate complex manual investigation processes at Amazon scale. You will work on some of the most fascinating challenges in applied AI—building systems that reason and act autonomously, learn rich representations from structured and relational data without extensive labels, adapt rapidly from limited examples, improve through feedback and interaction, seamlessly connect visual and textual understanding, and compress complex model capabilities into efficient, deployable systems. Your innovations will deliver significant impact to cost-of-serving customers while maintaining the highest standards of trust and safety.
This role offers end-to-end ownership—from initial research and proof-of-concept through production
deployment. You will see your innovations serving hundreds of millions of customers within months, not years.
Key job responsibilities
• Design and build expertise deep agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence with capabilities to handle feedback with long term as well as short term memory mechanisms.
• Design and build expertise agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence with capabilities to handle feedback with long term as well as short term memory mechanisms.
• Productionize large scale models built on top of SFT (Supervised Finetuning) and RFT (Reinforced fine tuning) approaches (GRPO with RLVR, Process/Outcome Reward Models), few shot approaches (Contrastive, Prototypical) based on multimodal datasets
• Enhance on existing Automatic prompt optimization techniques (GEPA & beyond) towards agentic optimization given the ground truth datasets to improve agentic planning.
• Build novel production ready Finetuned transformer architectures (using LORA/Q-LORA/LLM-JEPA etc) and conventional supervised & unsupervised ML solutions to aid the multiple potential automation requirements
• Identify customer and business problems at project level; invent or extend state-of-the-art approaches for complex LLM workflows involving unstructured text, documents, images, and relational data
• Author or co-author research papers for peer-reviewed venues; serve as PC member at conferences when aligned with business needs
• Prototype rapidly, iterate based on feedback, and deliver components at SDE 1+ level that integrate directly into production-scale systems
• Engineer efficient systems balancing model capability, deployment cost, and resource usage; write significant code demonstrating technical excellence and maintainability
• Scrutinize algorithm and software performance for improvements; resolve root causes leaving systems more maintainable
• Contribute to tactical and strategic planning—team goals, priorities, and roadmaps—while providing architectural guidance for AI systems
• Participate in engineering best practices with rigorous peer reviews; communicate design decisions clearly and participate in science reviews
• Train new teammates & interns on component construction and integration; mentor less experienced scientists and participate in hiring processes
About the team
Investigation technology Product team in TSE is responsible for the human-in-the-loop products and technology used in the risk investigations at Amazon. The team is also responsible for reducing the cost of performing the investigations, by automating wherever possible and optimizing the experience where manual interventions are needed. The team leverages state-of-the art technology and GenAI to deliver the products and associated goals.