3 days ago
Melbourne, AustraliaSenior
Responsibilities
- Advance planning, memory, skill acquisition, and multi-agent delegation architectures for production AI agents.
- Build reinforcement-learning infrastructure for customer-service models, including reward curricula, rollout systems, and production feedback loops.
- Develop multi-turn evaluation, automated trajectory analysis, quality gates, and CI-integrated regression detection.
- Design scalable guardrails for tool misuse, prompt injection, cascading actions, hallucination loops, access control, and output validation.
- Own both the science and systems required to improve agent performance and reduce inference costs.
Requirements
- 5+ years of experience building production ML/AI systems.
- Hands-on experience with agent architectures, including planning, tool dispatch, memory, and error recovery.
- Experience building internal evaluations and understanding the gap between public benchmarks and production performance.
- Fluency in Python and PyTorch.
- Familiarity with at least one agent framework and the judgment to build custom systems when appropriate.
- Optional depth in reinforcement learning for language models, including reward shaping, online/offline tradeoffs, and reward hacking.
Benefits
- Hybrid work arrangement with in-person collaboration at Zendesk offices and remote flexibility for part of the week.
- Inclusive workplace with equal-opportunity employment and reasonable-accommodation support.
Categories
About Zendesk
Zendesk builds a cloud-based customer service and contact center platform that unifies ticketing, messaging, voice, knowledge base, and AI-driven automation for businesses. It sells SaaS subscriptions and add-ons to companies across industries; founded in 2007 and headquartered in San Francisco, Zendesk is privately held under Hellman & Friedman after being taken private, and supports omnichannel CX and employee service workflows.