Senior Technical Program Manager, Human-in-the-loop Operations

Roblox Roblox · Consumer · San Mateo, CA · Technical Program Management

Senior Technical Program Manager to own the end-to-end lifecycle of Roblox's AI data labeling and model evaluation ecosystem. This role involves scaling data operations, managing a multi-million dollar budget and a distributed workforce, and driving the platform's evolution towards AI-assisted and LLM-accelerated operations. The goal is to build data infrastructure critical for the quality of Roblox's AI models across various applications.

What you'd actually do

  1. Lead data programs end-to-end. Own the full lifecycle of labeling and model evaluation workflows across Roblox's AI teams — from translating ambiguous ML requirements into structured annotation tasks, to overseeing contractor execution, quality review, and dataset delivery to model teams.
  2. Define and maintain ground truth. Develop and iterate on labeling guidelines, annotation schemas, and quality frameworks tailored to Roblox's unique data types: 3D mesh quality, texture coherence, search relevance, game novelty detection, NPC behavior evaluation, and AI-generated Luau code assessment.
  3. Manage a multi-vendor, distributed workforce. Oversee remote contractor teams, managing workforce allocation, productivity SLAs, quality calibration, and budget forecasting serving across multiple teams with Roblox.
  4. Drive the shift to AI-assisted workflows. Partner with Engineering to evaluate and implement LLM-based pre-labeling, LLM-as-a-Judge evaluation, and automated QA routing — reducing FTE coordination overhead and scaling throughput proportional to our annual volume growth.
  5. Partner cross-functionally across Roblox AI. Work closely with ML Engineers, Data Scientists, and Product Managers to translate model development needs into concrete, executable data programs, and communicate program status and quality trends to senior leadership.

Skills

Required

  • 5+ years of experience in Technical Program Management, Data Operations, or Operations Management within an AI/ML or data-intensive environment.
  • Deep hands-on experience with data labeling, annotation, or model evaluation — including designing annotation schemas, writing labeling guidelines, and managing quality control at scale.
  • Proven experience managing external vendor relationships and distributed contractor workforces, including workforce planning, quality oversight, and budget management.
  • Strong understanding of the ML lifecycle and the role of human-labeled data in training, fine-tuning, and evaluating models.
  • Proficiency in SQL and Python for querying datasets, analyzing label quality, and building operational dashboards. At a minimum, be able to use AI tools to generate queries.
  • Excellent written and verbal communication skills — able to write rigorous technical specifications and present complex data concepts to both ML researchers and business stakeholders.
  • Proven ability to drive cross-functional alignment and exercise strong judgment, knowing when to champion collaborative efforts versus leading independently.
  • Exceptional critical thinking skills with a demonstrated capacity to navigate ambiguity and execute effectively in 0-to-1 environments without structured guidance.
  • Solid grasp of software development life cycles and hands-on experience leveraging AI-assisted tools like Cursor, Claude, or Gemini to accelerate workflows.

Nice to have

  • Experience with LLM-based labeling pipelines, LLM-as-a-Judge evaluation, or human-AI calibration loops.
  • Familiarity with labeling platforms such as Label Studio, Scale AI, or Snorkel AI.
  • Experience evaluating 3D, games, videos, or multimodal data beyond standard text and image annotation.
  • Background in vendor platform evaluation or build vs. buy analysis.
  • Experience with data operations at a consumer platform operating at massive scale (100M+ users).
  • Knowledge of Roblox platform and its games, or general gaming experience.

What the JD emphasized

  • AI data labeling and model evaluation ecosystem
  • scale our data operations
  • AI-assisted, LLM-accelerated operations
  • build the data infrastructure that directly determines the quality of Roblox's AI models
  • human judgment is the critical ingredient for getting these models right
  • labeling guidelines
  • quality frameworks
  • ML requirements
  • contractor execution
  • quality review
  • dataset delivery
  • ground truth
  • annotation schemas
  • quality control at scale
  • ML lifecycle
  • human-labeled data
  • training
  • fine-tuning
  • evaluating models
  • label quality
  • operational dashboards
  • rigorous technical specifications
  • complex data concepts
  • ML researchers
  • business stakeholders
  • cross-functional alignment
  • strong judgment
  • collaborative efforts
  • critical thinking skills
  • navigate ambiguity
  • 0-to-1 environments
  • structured guidance
  • AI-assisted tools

Other signals

  • AI data labeling and model evaluation ecosystem
  • scale our data operations
  • driving the platform evolution from manual workflows to AI-assisted, LLM-accelerated operations
  • build the data infrastructure that directly determines the quality of Roblox's AI models