Carl Betzler
Currently building AI platform @ Nuro
I build the data pipelines, feedback loops, and eval systems that turn AI models into products that hold up in production. That work has spanned autonomous vehicles at Nuro and computer vision at scale in Amazon fulfillment centers.
Engineering teams needed a way to find specific scenarios across massive volumes of unstructured sensor data — without manual review.
Fast and precise embedding search — with natural language or image.
Data collection throughput was constrained by fleet scheduling, allocation, and sensor reliability.
Ops management tooling, automated troubleshooting, and route targeting.
Labeling was a bottleneck on both scale and consistency for training data.
Auto and LLM-based labeling infra, augmented with human judgement.
Manual inspection couldn't scale to catch product-quality issues across high-volume operations.
Deep learning models to identify anomalies and reduce manual review.