Loading...
Applied Scientist, Trustworthy Shopping Experience (TSE)
Amazon
On-sitemidPosted May 12, 2026
About the Role
Contribute to the design and development of agentic AI systems with multi-step reasoning and multimodal intelligence, productionize large-scale models, and build novel fine-tuned transformer architectures for automation requirements.
Requirements
Design and develop agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence, leveraging LLMs, reinforcement learning, and fine-tuned transformer architectures to automate complex manual investigation processes.
Full Job Description
Are you passionate about solving complex business problems at scale through Generative AI? Do you want to help build intelligent systems that reason, act, and learn from minimal supervision? 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 do this in close partnership with our selling partners, empowering them with best-in-class tools and expertise to offer a high-quality, compliant selection that customers trust.
As an Applied Scientist I, you will bring subject matter expertise in at least one relevant discipline (e.g., NLP, computer vision, representation learning, agentic architecture) to contribute to next-generation agentic AI solutions that automate complex manual investigation processes at Amazon scale. Working alongside senior scientists, you will map business goals—such as reducing cost-of-serving while maintaining trust and safety standards—to well-defined scientific problems and metrics. You will invent, refine, and experiment with solutions spanning agentic reasoning, self-supervised representation learning, few-shot adaptation, multimodal understanding, and model compression. With guidance from senior scientists, you will stay current on research trends and benchmark your results against the state of the art.
You will help design and execute experiments to identify optimal solutions, initiating the development and implementation of small components with team guidance. You will write secure, stable, testable, and well-documented production code at the level of an SDE I, rigorously evaluating models and quantifying performance. You will handle data in accordance with Amazon policies, troubleshoot issues to root cause, and ensure your work does not put the company at risk.
Your scope of influence will typically be at the self-level, with the possibility of mentoring interns. You will participate in team design and prioritization discussions, learn the business context behind TSE's products, and escalate problems with proposed solutions. You will publish internal technical reports and may contribute to peer-reviewed publications and external review activities when aligned with business needs.
This role offers a unique opportunity to contribute to end-to-end AI development—from research through production—with your contributions serving hundreds of millions of customers within months, not years.
Key job responsibilities
•Contribute to the design and development of agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence, including feedback and memory mechanisms, leveraging reinforcement learning techniques for agent decision-making and policy optimization, with input and guidance from senior scientists
• 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
•Contribute to building production-ready deep learning and conventional ML solutions, including multimodal fusion and cross-modal alignment techniques that seamlessly connect visual, textual, and relational understanding, to support automation requirements within your team's scope
•May co-author research papers for peer-reviewed internal and/or external venues, including contributions in areas such as multimodal representation learning and vision-language modeling, and contribute to the wider scientific community by reviewing research submissions, when aligned with business needs
•Prototype rapidly, iterate based on feedback, and deliver small components at SDE I level—including multimodal data pipelines and inference modules—that integrate into production-scale systems
•Write secure, stable, testable, maintainable, and well-documented code, balancing model capability, deployment cost, and resource usage across multimodal architectures while understanding state-of-the-art data structures, algorithms, and performance tradeoffs
•Rigorously test code and evaluate models across individual and combined modalities, quantifying their performance; troubleshoot issues, research root causes, and thoroughly resolve defects, leaving systems more maintainable
•Participate in team design, scoping, and prioritization discussions through clear verbal and written communication; seek to learn the business context, science, and engineering behind your team's products, including how multimodal signals contribute to trust and safety decisions
•Participate in engineering best practices with peer reviews; clearly document approaches and communicate design decisions; publish internal technical reports to institutionalize scientific learning
•Help train and mentor scientist interns; identify and escalate problems with proposed solutions, taking ownership or ensuring clear hand-off to the right owner
About the team
Trustworthy Shopping Experience 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 do this in close partnership with our selling partners, empowering them with best-in-class tools and expertise to offer a high-quality, compliant selection that customers trust.
As an Applied Scientist I, you will bring subject matter expertise in at least one relevant discipline (e.g., NLP, computer vision, representation learning, agentic architecture) to contribute to next-generation agentic AI solutions that automate complex manual investigation processes at Amazon scale. Working alongside senior scientists, you will map business goals—such as reducing cost-of-serving while maintaining trust and safety standards—to well-defined scientific problems and metrics. You will invent, refine, and experiment with solutions spanning agentic reasoning, self-supervised representation learning, few-shot adaptation, multimodal understanding, and model compression. With guidance from senior scientists, you will stay current on research trends and benchmark your results against the state of the art.
You will help design and execute experiments to identify optimal solutions, initiating the development and implementation of small components with team guidance. You will write secure, stable, testable, and well-documented production code at the level of an SDE I, rigorously evaluating models and quantifying performance. You will handle data in accordance with Amazon policies, troubleshoot issues to root cause, and ensure your work does not put the company at risk.
Your scope of influence will typically be at the self-level, with the possibility of mentoring interns. You will participate in team design and prioritization discussions, learn the business context behind TSE's products, and escalate problems with proposed solutions. You will publish internal technical reports and may contribute to peer-reviewed publications and external review activities when aligned with business needs.
This role offers a unique opportunity to contribute to end-to-end AI development—from research through production—with your contributions serving hundreds of millions of customers within months, not years.
Key job responsibilities
•Contribute to the design and development of agentic AI systems with multi-step reasoning, autonomous task execution, and multimodal intelligence, including feedback and memory mechanisms, leveraging reinforcement learning techniques for agent decision-making and policy optimization, with input and guidance from senior scientists
• 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
•Contribute to building production-ready deep learning and conventional ML solutions, including multimodal fusion and cross-modal alignment techniques that seamlessly connect visual, textual, and relational understanding, to support automation requirements within your team's scope
•May co-author research papers for peer-reviewed internal and/or external venues, including contributions in areas such as multimodal representation learning and vision-language modeling, and contribute to the wider scientific community by reviewing research submissions, when aligned with business needs
•Prototype rapidly, iterate based on feedback, and deliver small components at SDE I level—including multimodal data pipelines and inference modules—that integrate into production-scale systems
•Write secure, stable, testable, maintainable, and well-documented code, balancing model capability, deployment cost, and resource usage across multimodal architectures while understanding state-of-the-art data structures, algorithms, and performance tradeoffs
•Rigorously test code and evaluate models across individual and combined modalities, quantifying their performance; troubleshoot issues, research root causes, and thoroughly resolve defects, leaving systems more maintainable
•Participate in team design, scoping, and prioritization discussions through clear verbal and written communication; seek to learn the business context, science, and engineering behind your team's products, including how multimodal signals contribute to trust and safety decisions
•Participate in engineering best practices with peer reviews; clearly document approaches and communicate design decisions; publish internal technical reports to institutionalize scientific learning
•Help train and mentor scientist interns; identify and escalate problems with proposed solutions, taking ownership or ensuring clear hand-off to the right owner
About the team
Trustworthy Shopping Experience 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.