Model Deployment Platform Engineer at Cruise

Developing tooling to assist in debugging the model deployment process, going from research models to optimized, production models.

At Cruise, I was on the Model Deployment Platform team (MDP). Specifically, I was responsible for creating Python tooling that we would use to debug models going through deployment, from PyTorch to (usually) TensorRT. In deployment, we’d often get numerical disparity between the optimized model and the original PyTorch model. My tools helped debug these issues, quantifying the errors and making it easy to find the root-cause. The project was fun in that I got to build for machine learning engineers, customers that I understand. I also really liked it when someone would message me on Slack telling me my new tool saved them hours while debugging their model. :)

What I Learned

• Model deployment workflows, how a large company deployed ML models.
• Iterating on tooling based on customer feedback.
• Software development on an agile team.