Founding ML Engineer

a16z-speedrun•San Francisco, CA

About The Position

Own the loop from raw robot experience to a model running on real hardware. Munari observes what robots see, sense, decide, and do. The role uses this data to detect abnormal behavior, understand failures, evaluate policies, identify recurring failure modes, and help robotics teams improve their systems. This role sits between research engineering and production machine learning. Work may include training multimodal models, redesigning datasets, optimizing inference, deploying models to edge devices, and testing models against real robots. The focus is on research that becomes a working system.

Requirements

  • Experience training modern vision, transformer, diffusion, generative, or multimodal models.
  • High productivity with PyTorch, JAX, or a comparable machine-learning stack.
  • Understanding of both model behavior and the systems needed to train and deploy models reliably.
  • Ability to work comfortably across research code, data pipelines, production services, and edge deployment.
  • Excitement about messy real-world data and willingness to begin without a perfectly clean benchmark.
  • Strong experimental judgment and ability to distinguish genuine improvement from an attractive demo.

Nice To Haves

  • Experience with distributed training, large datasets, inference optimization, or GPU performance.
  • Experience with imitation learning, reinforcement learning, diffusion policies, VLAs, world models, sim-to-real, or active learning.
  • Experience with TensorRT, ONNX, Triton, CUDA, Isaac Sim, MuJoCo, or ManiSkill.
  • Robotics experience; exceptional candidates from vision, video, autonomous systems, or large-scale generative modeling are also encouraged to apply.

Responsibilities

  • Train multimodal models using video, language, proprioception, motor currents, force, tactile, IMU, telemetry, trajectories, and policy traces.
  • Build models for anomaly detection, episode scoring, failure classification, incident understanding, representation learning, and behavioral evaluation.
  • Integrate, evaluate, adapt, or fine-tune vision-language-action models and other robot policies across different embodiments.
  • Develop training and evaluation methods for limited, noisy, heterogeneous, and highly imbalanced real-world robot data.
  • Convert failures, operator takeovers, recoveries, and successful episodes into evaluations, labeled datasets, and training signals.
  • Distill, quantize, prune, compile, and optimize models for deployment on constrained edge hardware.
  • Design benchmarks, ablations, fault-injection experiments, and metrics to determine whether a model will improve real-world operation.
  • Work with the infrastructure team on distributed training and data systems, and with the edge team on low-latency inference.
  • Validate ideas in simulation and close the loop on physical robots.
© 2026 Teal Labs, Inc
Privacy PolicyTerms of Service