Atlassian is seeking a Principal Machine Learning Engineer to set the technical direction for AI agent quality in Studio. This is a hands-on technical leadership role. You'll define how we measure whether agents built on our platform are doing good work, and lead the work to make them better.
Your future team
Studio is where Atlassian customers build, run and manage AI agents under the Rovo brand. Rovo Agents is a model-agnostic cloud agent platform. Teams give an agent instructions, knowledge, tools and skills, then put it to work in Jira and Confluence, inside automation rules, or as an autonomous agent that works on its own.
Because customers build these agents themselves, quality can't be checked once before release. Every agent has its own instructions, data and tools, and any of them can be misconfigured or connected to a tool that fails. We ask three questions of every agent: did it understand and follow the task, did it use its tools correctly and report honestly what it did, and did the result actually help?
The team is software and machine learning engineers based in Sydney and Melbourne, with remote options. We work closely with product, data science and Atlassian's central AI teams. This is applied ML on real agentic systems in production.
What you'll do
Own the agent quality architecture. Define what we measure, how production traffic is sampled, how it is judged, and how results roll up without hiding important failure modes behind one score.
Design for cost and accuracy. Decide which quality questions can be answered with deterministic checks, which need behavioral heuristics, and which need contextual model-based judgment. Build the evaluation system around those decisions.
Architect evaluation and experimentation. Define the benchmarks, datasets, evaluation environments and observability needed to assess agent behavior and task success. Guide teams in building reproducible experiments and calibrated evaluators that inform model, prompt and launch decisions.
Make quality part of how we ship. Set up regression gates and pre-release checks for agent, prompt and model changes, and drift detection on production traffic.
Lead the hardest quality problems hands-on. Examples include agent selection and routing, tool-use reliability, and quality for agents that run inside automations with no user present to give feedback.
Lead technical direction across teams. Set a multi-quarter roadmap with product and data science, partner with central AI teams and model providers, and take accountability for measurable improvements in agent quality.
Grow ML craft. Mentor engineers through design reviews and hands-on work in a team made up mostly of software engineers.
Your background
On the first day, we'll expect you to have:
6+ years in ML/AI engineering, with deep expertise in LLMs, NLP, information retrieval or recommendation.
A proven track record of building and shipping LLM-powered or agentic systems used by real customers, and improving them over time through measurement.
Strong evaluation and experimental skills. You can design useful benchmarks, assess label and judge quality, recognize leakage and confounding factors, and connect offline results to product impact.
Strong software and data engineering skills, with the ability to validate designs in code and work deeply with data pipelines, model serving and production observability.
Strong product judgment: you can find the real problem, define measurable outcomes, and choose between deterministic, classical ML and LLM-based approaches.
A record of influencing technical direction across multiple teams and mentoring senior engineers.
Regular use of AI tools in your own work, and the ability to turn new research into production results.
Sound judgment about trade-offs across quality, latency, cost, reliability, privacy and access control.
It's great, but not required, if you have:
Experience building agent platforms where customers configure the agents.
Experience with trajectory-level evaluation of multi-step agents, or prompt and agent optimization.
Experience in enterprise SaaS.
A master's or PhD in Computer Science, Machine Learning or a related field.
