Applied Scientist

About the Role

The Alexa Sensitive Content Intelligence (ASCI) team is developing Responsible AI (RAI) solutions for Alexa+, a next-generation, AI-powered virtual assistant. The mission is to build AI safety systems that protect millions of Alexa customers daily by ensuring LLMs provide safe, trustworthy responses, understanding nuanced human values across cultures, and maintaining customer trust at scale.

What You'll Build

• Pioneer breakthrough solutions in Responsible AI at Amazon's scale.
• Train models that set new safety standards.
• Design automated testing systems to hunt for vulnerabilities.
• Certify systems powering millions of daily conversations.
• Create intelligent evaluation systems that judge responses with human-level insight.
• Build models that understand what makes interactions safe and delightful.
• Craft feedback mechanisms for Alexa+ to grasp complex customer conversations.
• Build AI agents for real-time detection and fixing of production issues.
• Innovations will shape how Alexa+ learns, evolves, and earns customer trust.
• Shape how millions of people interact with AI confidently through novel safety mechanisms and sophisticated evaluation techniques.
• Define industry standards for responsible AI.
• Train models that remain truthful and grounded across diverse contexts.
• Build reward models that capture the nuanced spectrum of human values across cultures and languages.
• Create automated systems that continuously discover and address potential issues before customers encounter them.
• Collaborate with world-class scientists, product managers, and engineers to transform state-of-the-art ideas into production systems serving millions.

Key Responsibilities

• Contribute to end-to-end Responsible AI solutions.
• Develop and refine algorithms that detect and eliminate false information.
• Build models that identify vulnerabilities exploited by bad actors.
• Train systems that understand human values across diverse cultures.
• Apply state-of-the-art NLP and computer vision techniques to content classification and safety challenges.
• Turn cutting-edge research into production systems that protect millions of customers daily.
• Contribute to identifying, defining, and mitigating AI safety risks early.
• Build deep expertise in responsible AI.
• Work within multi-agent architectures, implementing components where distributed AI systems maintain human oversight and control.
• Build models and run experiments, from consensus-driven reasoning modules to orchestration components, to strengthen trustworthy AI at scale.
• Lead efforts to identify, define, and mitigate risks early, protecting customers before they're impacted.
• Mentor the next generation of AI safety experts.
• Design multi-agent architectures where distributed AI systems exhibit emergent collective intelligence while maintaining human oversight and control.
• Drive innovations from consensus-driven reasoning frameworks to sophisticated orchestration patterns, setting new standards for trustworthy AI.
• Experiment with breakthrough techniques that push the boundaries of AI safety.
• Partner with engineering to transform research into production-ready solutions.
• Tackle complex optimization challenges.
• Brainstorm with Product teams, translating ambitious visions into concrete objectives.
• Shape critical deployment decisions by reviewing impactful work and guiding go/no-go calls.
• Mentor the next generation of AI safety leaders.

About The Team

• Pioneers Responsible AI for conversational assistants.
• Ensures Alexa delivers safe, trustworthy experiences across all devices, modalities, and languages worldwide.
• Works on frontier AI safety challenges.

Requirements & Tech Stack

• Deep expertise in state-of-the-art NLP, Large Language Models.
• Track record of building scalable ML systems.
• Experience programming in Java, C++, Python or related language.
• Experience in state-of-the-art deep learning models architecture design, deep learning training and optimization, and model pruning.

Qualifications

• Experience building machine learning models or developing algorithms for business application.
• Experience in patents or publications at top-tier peer-reviewed conferences or journals.
• Experience using Unix/Linux.
• Experience with programming languages such as Python, Java, C++.