Software Development Engineer, Amazon
Amsterdam, Netherlands · November 2021 – November 2023
Machine learning libraries and computer vision for Ring devices, including motion detection shipped in cameras with 1.5M combined sales.
- Machine learning
- Computer vision
- Edge devices
- AWS
- Low-level optimizations
Impact
- Designed and built an ML conversion library for ONNX-based models, with embedded C++ and SIMD-level optimizations, that performed 40% better than external alternatives.
- Integrated motion detection ML models into Ring Indoor Cam, Stick Up Cam Pro, and Spotlight Cam Pro (1.5M units sold combined), including business logic, model porting, and feature rollout.
- Built an AWS-hosted auto-documenting service that streamlined embedding and maintenance of computer vision algorithms across multiple product lines.
Ring devices run vision models on constrained hardware, where accuracy, latency, and power draw trade off directly against each other.
At Amazon I worked on machine learning libraries and computer vision for these devices. I designed an ML conversion library for ONNX-based models in embedded C++ with SIMD-level optimizations, then used it to bring motion detection to Ring Indoor Cam, Stick Up Cam Pro, and Spotlight Cam Pro. I also built an AWS-hosted service that auto-documents ML model integrations, so computer vision algorithms are easier to embed and maintain across product lines.