Machine Learning Performance Engineer, Covariant

• Obsession with building products that deliver value to customers through solving hard ML and robotics challenges
• 3+ years of professional experience working with real-world software systems, AI products, or AI research
• Proficiency in Python and experience with tensor libraries (e.g. NumPy, PyTorch, TensorFlow)
• Working knowledge of Linux, SQL, and web (e.g. HTML, Javascript, etc.)
• A solid mathematical and statistical foundation with an understanding of how to apply ML concepts (e.g. training, evaluation, testing, fine-tuning, data sampling, etc.)
• Demonstrated strong problem-solving ability: analyzing real-world problems and formulating solutions, iterating and formulating, shipping, and making an impact on products for customers
• Clear communication and collaboration across teams: taking requests from customers, and PMs, and prioritizing work across SW and HW teams

Desirable

• Each ML Performance Engineer will be expected to own a product family and use their understanding of the Covariant Brain to make improvements toward making robots highly autonomous
• Own all performance aspects of a robotic application
• Understand and iterate on all aspects of the Brain (models, algorithms, tooling, etc.) to resolve production failures and deliver human-level autonomy
• Analyze robot performance for different errors, understand their root causes, and test/deploy improvements to production
• Build tools to analyze a highly complex system of robots, models, components, and sensors to understand errors and enable more people across the org to make improvements to the system
• Take real-world challenges and engineer ML & robotics solutions by training, testing, tuning, iterating, and deploying new and existing models to solve production problems
• Collaborate closely with research, SW, HW, and infrastructure teams to deliver highly reliable robotic applications
• Work closely with the HW team on the design of a robotic application including the gripper, vision suite, and station layout
• Identify throughput bottlenecks in the application and work with the SW team to pinpoint and resolve them
• Take experimental models from the research team, test them to validate improvement, and deploy them to solve application problems
• Travel to the customer site as needed to resolve issues with robots in production (not expected for more than once per quarter, if needed at all)