Senior Machine Learning Engineer

<h1><strong>About the Team & Role:</strong></h1> <p>We're building <strong>SAGE</strong>, Rubrik's Semantic AI Governance Engine, which is the first system designed to monitor, govern, and remediate autonomous AI agents in real time. SAGE powers Rubrik Agent Cloud: enterprises define governance policies in natural language, and SAGE's custom small language models act as judges on every agent action. These models are fast enough to sit in the live request path and accurate enough that customers trust them with allow/block decisions on production traffic.</p> <p>At its core, SAGE is "LLM-as-judge" applied to AI governance, utilizing the same technique most teams use for offline evaluation but productionized for real-time enforcement at enterprise scale. Our first-generation SLM Policy Guard already outperforms the larger frontier models we've benchmarked against on accuracy while running approximately 5x faster on the same workload. We're hiring to push that lead even further.</p> <p>As an Applied ML Engineer on the SAGE team, you'll work end-to-end across the model lifecycle: curating data, training small models, serving them at production latency, and closing the feedback loop with real customer signals. The models you build don't just enforce policies in the live request path; they will also drive Agent Rewind, Rubrik's capability to instantly and precisely undo destructive autonomous-agent actions and restore the affected data to a trusted state. </p> <p>We're a collaborative, applied team that ships models to enterprise customers within weeks, and we're passionate about proving that small, specialized models can outperform frontier LLMs at the problems that matter most for AI safety and governance.</p> <h2><strong>Nature of the Specialized Duties</strong></h2> <h3>➢ <strong>Training, Fine-Tuning, and Distilling Production Small Language Models and Classifiers (25% of time)</strong></h3> <ul> <li>Owning the full training lifecycle for the SLMs and classifiers in SAGE's real-time enforcement path, including base-model selection, supervised fine-tuning, preference optimization (DPO/RLAIF), and distillation from frontier teacher models.</li> <li>Training anomaly and action-severity models that catch novel agent-side attack patterns at real-time decision latency, such as supply-chain compromises or emergent destructive behaviors not covered by any explicit policy. Severity scores route the highest-impact events to Agent Rewind for precise remediation.</li> <li>Designing adversarial training pipelines like purpose-built adversarial agents and automated red-teams whose outputs feed directly into the next training run, turning every discovered weakness into a permanent model improvement.</li> <li>Pushing