At Google, I led work on on-device ML models that mitigate GNSS multipath reflections in urban environments in real time. The new model reduced wrong-side-of-the-street errors on Android phones by 30%+.
I owned the ML lifecycle from feature engineering and training through distillation and quantization, with inference designed to run in milliseconds within a strict battery budget. I also developed Google Cloud Dataflow workflows for continuous model training and evaluation.
At Cloudflight GmbH (previously Catalysts), I architected and deployed an object-tracking system for edge devices with over 95% accuracy across dozens of sites. I also developed visual search and synthetic-data pipelines for industrial applications.
