Applied Scientist - Multimodal AI for Customer Experience

Amazon Inc.; Berwick (Columbia County); On-site; USD 143,000 - 193,000; Full time

Job summary

Amazon's CXBT team is seeking an Applied Scientist to advance agent-based modeling for customer experience optimization. You will research state-of-the-art techniques, lead scientific innovation, and build foundational agentic simulation capabilities that integrate multimodal inputs and real-time knowledge retrieval.

You will design frameworks for multi-modal content handling, develop user modeling and behavioral simulation, curate diverse datasets with privacy in mind, and publish robust

Qualifications

• 3+ years building models for business applications.
• PhD or equivalent research experience.
• Experience programming in Java, C++, Python or related language.
• Experience in algorithms, data structures, parsing, numerical optimization, data mining, parallel and distributed computing, HPC.
• Publications or patents in top-tier conferences or journals.

Responsibilities

• Design and implement multi-modal content handling frameworks for diverse agents.
• Research user modeling and behavioral simulation to build agentic frameworks.
• Develop novel algorithms and modeling techniques.
• Acquire and curate diverse datasets while preserving user privacy.
• Create robust evaluation metrics and test sets for language model performance.
• Apply responsible AI practices throughout the development process.
• Document methodologies and design decisions clearly.

Skills

Java

C++

Python

Machine learning

Algorithms

Education

PhD or equivalent

Job description

Amazon's CXBT team is seeking an Applied Scientist to advance agent-based modeling for customer experience optimization. You will research state-of-the-art techniques, lead scientific innovation, and build foundational agentic simulation capabilities that integrate multimodal inputs and real-time knowledge retrieval.

You will design frameworks for multi-modal content handling, develop user modeling and behavioral simulation, curate diverse datasets with privacy in mind, and publish robust