ML Engineer, Agentic Labeling
Posted 6 months ago San Francisco, CA full-time
About This Role:
Build the ML models that power Avala's agentic labeling system — the pipeline where AI models generate initial annotations, human experts refine them, and the feedback loop continuously improves both. You'll work at the intersection of foundation models, active learning, and production annotation workflows.
What You'll Do:
Develop and fine-tune foundation models for auto-labeling (2D detection, 3D cuboids, segmentation)
Build active learning systems that intelligently route tasks between AI and human annotators
Design feedback loops that use human corrections to improve model accuracy over time
Implement model evaluation pipelines that measure annotation quality and coverage
Scale inference infrastructure to handle real-time auto-labeling across sensor modalities
Collaborate with operations teams to optimize the human-AI annotation workflow
Requirements:
3+ years training and deploying ML models in production
Strong experience with object detection, segmentation, or 3D perception models
Proficiency in Python, PyTorch, and distributed training frameworks
Experience with active learning, semi-supervised learning, or human-in-the-loop ML
Understanding of ML ops — model versioning, A/B testing, monitoring
Nice to Have:
Experience with LiDAR perception or multi-sensor fusion models
Familiarity with foundation models (SAM, DINO, or similar) for annotation
Published research in active learning or interactive annotation
What We Offer:
Competitive salary and meaningful equity in the team building Fleet Learning Infrastructure for Physical AI
Full health, dental, and vision coverage, plus FSA and unlimited PTO
Work on the closed-loop data infrastructure that the next generation of autonomous systems learns from
A team that built Tesla Autopilot's data infrastructure, now building something bigger