We are looking for a passionate, talented, and inventive Data Scientist with a strong machine learning and analytics background to help build industry-leading language technology powering Rufus, our AI-driven search and shopping assistant, helping customers with their shopping tasks at every step of their shopping journey. This innovative role focuses on developing and optimizing large language model (LLM)-powered conversational experiences. The core emphasis is to get the best performance out of state-of-the-art LLMs via careful and methodical instruction design, contextual grounding, informed choices of MCP tools and agent/multi-agent systems, evaluation frameworks, and experimentation to systematically improve LLM quality, robustness, and customer impact. The work combines scientific rigor with product intuition to systematically raise the bar for conversational AI performance at Amazon scale. Our mission in conversational shopping is to make it easy for customers to find and discover the best products to meet their needs by helping with their product research, providing comparisons and recommendations, answering product questions, enabling shopping directly from images or videos, providing visual inspiration, and more. We do this by leveraging advanced analytics, Natural Language Processing (NLP), Machine Learning (ML), A/B testing, causal inference, and data-driven insights to continuously improve our systems. Key job responsibilities As a Data Scientist on our team, you will develop and maintain LLM instructions iterations and evaluation frameworks, including automated eval pipelines, LLM-as-a-judge methodologies, rubric design, and dataset curation to measure nuanced aspects of response quality. You will partner with the wider org to experiment with techniques such as retrieval augmentation, context enrichment, prompt decomposition, and model fine-tuning or post-training strategies, if and when applicable. You will leverage petabytes of data and identify opportunities to leverage machine learning models aimed at making conversational systems more performant. A day in the life You will: Perform hands-on analysis of large-scale multimodal interaction datasets to develop insights into how customers engage with conversational AI systems and how to improve response quality and customer experience. Use statistical methods, experimentation, and data-driven analysis to develop scalable approaches for measuring, evaluating, and optimizing large language model (LLM)-based shopping assistant systems, leveraging structured and unstructured contextual signals. Design and analyze A/B tests and experiments to evaluate new features and model improvements, ensuring statistical rigor and actionable insights. Develop metrics, dashboards, and reporting frameworks to monitor system performance, customer engagement, and business impact. Conduct deep-dive analyses to identify opportunities for improving conversational relevance, grounding, customer satisfaction, and downstream business impact. Collaborate with Applied Scientists and Engineers to translate analytical insights into production systems, working closely on model evaluation and deployment. Establish automated processes for large-scale data analysis, ETL pipelines, metric generation, and experimentation frameworks. Communicate results and insights to both technical and non-technical audiences, including through presentations, written reports, and data visualizations. About the team The Rufus Features Science team, based in London, works alongside ~150 engineers, designers and product managers, shaping the future of AI-driven shopping experiences at Amazon. The team works on every aspect of the Rufus AI, from making Rufus agentic, enabling customers to set price alerts or empower Rufus to act on their behalf and automatically purchase products when the price is right, to understanding multimodal user queries and generating answers that combine text, image, audio and video, including deep research reports that scour the web and the Amazon catalog to provide detailed and personalised shopping guidance. We utilize and advance state-of-art techniques in the fields of Natural Language Processing, gen AI, Information Retrieval, Machine/Deep Learning, and Data Mining. We validate our work by actively participating in the internal and external scientific communities.
