Principal Machine Learning Engineer, Engineering Efficiency

Why Engineering Efficiency? 

The Engineering Efficiency AI Infrastructure Pod acts as Roblox’s center of excellence for applying AI to software engineering. We develop the Code Intelligence components of the Builder AI Platform, enabling first-party agents to automate engineering toil such as refactoring and unit test generation

At Roblox, we aren't just trying to build another autocomplete tool or a basic LLM wrapper; we are architecting a Code Intelligence platform. Our mission is to move beyond simple code suggestions and into the realm of "agentic reasoning", creating a system that understands the deep, multi-dimensional context of a massive, 20-year-old engine and can autonomously handle the mechanical toil of modern development.

As AI makes it possible to generate code at an unprecedented velocity, our team’s focus shifts from writing code to maintaining quality, safety, and performance at a planetary scale. We are building the reasoning layer and the "Human-in-the-Loop" (HITL) gates that ensure this AI-driven evolution doesn't compromise the stability of a platform serving 100M+ daily active users. We aren't just using the industry standard; we are inventing the new one.

Joining this team means you will be a leader, bridging the gap between our AI teams, Eng Efficiency and Feature devolvement teams. We are looking for a visionary who can build platforms and first-party agents that don't just suggest code, but actually understand how to refactor, tune, and evolve the Roblox ecosystem. 

You Will: 

• Architect the Reasoning Layer and design systems that allow agents to navigate a billion-line codebase with high precision.
• AI Workload Optimization: Ensure our AI infrastructure is performant and cost-effective as it scales to become our primary compute driver.
• Lead New Techniques: Develop new methods for synthetic data generation and agent evaluations that outperform current industry benchmarks.

You Have: 

• Beyond "Off-the-Shelf": We are looking for an expert who can move past basic LoRA or full-parameter fine-tuning to implement Reinforcement Learning from Compiler Feedback (RLCF) and Domain-Specific Distillation.
• Code-Specific Optimization: The goal is to fine-tune models to understand the unique constraints of the Roblox Luau language, our internal APIs, and our proprietary high-performance engine architecture.
• Continuous Learning Loops: Architecting the infrastructure that allows our models to "learn" from every human-corrected diff, ensuring our internal models stay ahead of general-purpose industry standards.

You Are:

• Comfortable with Ambiguity: You thrive in undefined or open-ended problem spaces, providing structure, clarity, and decisive direction to your teams.
• A Pragmatic Builder: You are scrappy and impact-oriented. You view undefined data and messy systems as opportunities to build structure rather than blockers to progress.
• An Inspiring Leader: Passionate about developing the next generation of technical leaders, managers, and engineers.
• An Executive Communicator: Highly effective at communicating complex technical concepts to both engineering teams and non-technical executive leadership.
• Data & System Oriented: You understand that robust data and systems are the foundation of any production application, and you design infrastructure for scale, correctness, and reliability.
• Curious & Creative: You enjoy tackling hard problems, exploring new technologies, and driving continuous improvements in both systems and workflows.