AI Scientist - (AI Verticals)
What You Can Expect
In this role, you will tackle one of the most compelling challenges in production AI: enabling agents to continuously improve without compromising customer data privacy. You will own the development of synthetic data pipelines, advanced evaluation frameworks, and privacy-preserving feedback loops that drive measurable improvements in Zoom Virtual Agent performance, with your research shipping directly to production and serving real customers across Zoom’s growing portfolio of AI-powered products. As a part of the AI Verticals organization, you will operate with high autonomy in a fast-moving, research-to-production environment where innovations reach customers significantly faster than traditional research timelines.
This is a hybrid position that requires on-site presence four (4) days per week at our San Jose, CA or Bellevue, WA office. Remote arrangements are not available for this role.
About the Team
The AI Verticals team sits at the intersection of deep research and real-world production impact, advancing the intelligence and continuous improvement of Zoom’s AI agent products. Our work spans agent evaluation, model optimization, synthetic data generation, and self-improving systems, with direct influence on Zoom Virtual Agent and other products across the Zoom portfolio. The team operates a global development model with engineering partners across time zones, enabling research innovations to reach customers faster than traditional research organization timelines.
Responsibilities
Research and develop novel learning mechanisms that enable AI agents to continuously improve while adhering to customer data privacy and compliance requirements.
Develop synthetic data generation and simulation approaches that capture real production error patterns and customer behavior at scale, enabling model improvement through simulated scenarios.
Design advanced evaluation frameworks that go beyond traditional test sets and LLM-as-judge approaches, addressing how to assess and improve agent quality at scale without compromising customer data privacy.
Build scalable production feedback loops that leverage privacy-preserving signals and simulated scenarios to drive continuous improvements in agent performance.
Analyze customer needs and business outcomes, including resolution rate, escalation patterns, and conversation quality, to guide research priorities and deliver measurable improvements in agent performance.
Collaborate with engineering teams to deploy research innovations into production systems, balancing scientific rigor with practical engineering constraints.
Contribute to the research community by publishing findings on self-improving agents, synthetic data generation, and privacy-preserving learning techniques.
What We’re Looking For
Master’s degree or higher in Computer Science, Machine Learning, Artificial Intelligence, or a related field with a research focus on autonomous agents, reinforcement learning, self-supervised learning, or LLM-based systems, or equivalent practical experience in these areas.
Deep hands-on expertise with large language models (LLMs) applied to agent systems or real-world production scenarios.
Strong foundation in machine learning and reinforcement learning, with demonstrated experience in self-improving systems, continual learning, or agents that learn from limited feedback.
Creative approach to evaluation challenges, with the ability to develop novel metrics and frameworks for agent quality in privacy-constrained environments, beyond standard test sets and LLM-as-judge methods.
Programming proficiency in Python, C++, or Java, with hands-on experience in AI frameworks such as PyTorch or TensorFlow.
Demonstrated ability to translate research innovations into production systems at scale, balancing scientific rigor with practical engineering timelines.
Intellectual curiosity, bias for action, and growth mindset, with the drive to take ownership of…