Software Engineer Intern - ML Systems

Apptronik • Austin, TX

4d • Onsite

About The Position

Apptronik is seeking a Software Engineering Intern to join our ML Ops team for a 12-week fall project. In this role, you will work at the intersection of robotics and applied machine learning — building data annotation tooling and optimizing ML models that run on humanoid hardware. You will help close the loop between raw robot experience data and deployable, hardware-ready models for Apollo, Apptronik’s humanoid robot. You will take ownership of two interconnected workstreams: (1) building or extending data annotation tools that let the team efficiently label and curate robot experience data, and (2) applying ML model optimization techniques — quantization, distillation, and inference profiling — to improve the throughput and efficiency of models deployed on physical systems. You will work alongside the simulation engineering, data platform, and learning teams, and contribute directly to how Apptronik turns ML research into production robot behavior.

Requirements

• Demonstrated ability to write clean, tested, maintainable code for ML tooling, data pipelines, and automation.
• Comfortable in a Linux environment; competence with Git, Docker, and modern Python tooling (pytest, uv/poetry, type hints).
• Hands-on experience with PyTorch or similar; familiarity with model quantization (INT8/FP16), ONNX export, or TensorRT is a plus.
• Coursework or project experience with robotic systems — kinematics, control, sensors, or simulation (ROS, MuJoCo, Isaac Sim, Gazebo, or comparable).
• Experience moving data between annotation, training, evaluation, and storage stages.

Nice To Haves

• Prior experience with data annotation workflows, labeling interfaces (Label Studio, CVAT, custom tooling), or human-in-the-loop data pipelines.
• Familiarity with RL training loops, policy rollouts, or vision-language-action (VLA) models.

Responsibilities

• Design and implement tooling for efficient annotation and curation of robot experience data — including sensor observations, trajectories, and task outcomes — in formats compatible with the team’s data lake (MCAP, S3/MinIO).
• Profile, quantize, and/or distill ML models (RL policies, VLA controllers, or action heads) to reduce inference latency and memory footprint for deployment on robot hardware.
• Build evaluation harnesses that benchmark optimized model performance against baseline, tracking metrics relevant to physical deployment (latency, memory, task success rate).
• Connect annotation outputs and optimized model artifacts with the team’s existing artifact storage (S3/MinIO), training pipelines, and Kubernetes-based execution environment.
• Produce design docs, runbooks, and example configurations so tooling can be adopted by controls, learning, and data platform teams after the internship.

What This Job Offers

• Job Type: Full-time
• Career Level: Intern
• Education Level: No Education Listed