Jan Rudy
It's me!
I build ML systems for robots, vision, generative models, and other things that need to work in the real world.
Most of my work lives in the space between research and engineering. I like taking a good idea, turning it into something testable, and then doing the less glamorous work required to make it reliable.
Lately that has meant robotic manipulation at warehouse scale: helping robots understand messy scenes, predict what is likely to happen next, and make better decisions across large deployed fleets.
I care about models that are useful in the real world, but also about the machinery around them: clean data, reproducible training, reviewable experiments, offline metrics that actually predict production behaviour, and careful rollouts.
Before robotics, I worked on generative models and representation learning. Before machine learning, I worked as a re-recording mixer for film and television. That's kinda weird.
TL;DR
Interests
- Deep learning for robotics
- Computer vision and perception
- Generative models and controllable generation
- Learned robot behaviour and manipulation policies
- Simulation-informed deployment
- Offline evaluation and production A/B testing
- ML training systems, reproducibility, and reviewability
- Research-to-production engineering
Selected Experience
-
Robotic manipulation at scale
Built and shipped machine learning systems for robotic picking in messy warehouse environments, spanning perception, manipulation, task-success prediction, simulation validation, deployment, and production experiments across large robot fleets.
-
Learning from robot outcomes
Built models that predict downstream robot outcomes like pick success, scan success, and place success. A lot of this work comes down to making offline metrics honest enough that model improvements show up in production.
-
Research systems that scale
Developed distributed training workflows, dataset pipelines, experiment tracking, and ML configuration tooling so research could move faster without becoming impossible to reproduce or review.