Zendesk
Zendesk
Zendesk develops customer service and sales software that helps organizations manage conversations, support requests, and customer relationships in one connected environment. Its platform brings together messaging, live chat, AI-powered automation, data privacy and protection tools, and an agent workspace, giving teams flexible ways to organize service operations as they grow. Zendesk also offers professional services and training resources, with a stated focus on personalized customer interactions, integrated workflows, diversity and inclusion, and sustainability.

Senior Machine Learning Engineer, AI Agents

Advance AI agent planning, memory, evaluation, and safety for Zendesk’s customer service resolution platform. Build reinforcement learning systems using production trajectories.

Description

  • Design memory structures and autonomous learning methods for AI agents.
  • Improve agent architecture through ambiguous goal decomposition, memory tier management, selective skill acquisition, and A2A-based multi-agent delegation.
  • Build customer service agent models with reinforcement learning on production trajectories.
  • Lead reinforcement learning infrastructure, including reward curricula, rollout systems, and feedback loops.
  • Develop multi-turn evaluations and automated trajectory analysis.
  • Add CI quality gates and regression detection to prevent deployments when agent performance declines.
  • Implement enterprise guardrails with supervisor patterns, capability-based access controls, and output validation.
  • Collaborate with colleagues in Zendesk’s hybrid work environment.

Requirements

  • At least five years of experience building production machine learning or AI systems.
  • Practical experience with agent architectures, including planning, tool dispatch, memory, and error recovery.
  • Strong evaluation judgment and experience creating internal evaluations.
  • Fluency in Python and PyTorch.
  • Familiarity with at least one agent framework.
  • Sound judgment about when to build custom solutions.
  • Reinforcement learning experience or substantial expertise with language models is optional; relevant areas include reward shaping, online and offline tradeoffs, and using reward hacking as a diagnostic signal.

Benefits

  • Hybrid work with the flexibility to work remotely for part of the week.
  • An inclusive workplace.
  • Reasonable accommodations for applicants with disabilities and disabled veterans.

Related Jobs