Building & deploying machine learning solutions to prevent theft from pickup trucks.• Currently working on multi-object radar track classification (multi-variate time series classification)• Core developer & maintainer of our computer vision library which runs on our first product, the Pickup Cam (C++ development, unit/integration testing, debugging, deployment, automated CI/CD with Gitlab, model conversion/quantization/integration)• Involved in the full ML lifecycle for our CV models: data collection, annotation, QA, training, evaluation, deployment, and monitoring• Defined data collection campaigns, wrote objective annotation guidelines, and worked with third-party annotation companies• Demonstrated the ability to use 3D computer vision for advanced threat detection via a sensor fusion proof of concept (RADAR + Camera)• Prototyped and evaluated different hardware solutions (RADARs, MCUs, ...) and their associated software solutions• Interviewed candidates for various roles within the AI team

• Applied a mixture of concepts such as semantic segmentation, object detection, and 3D computer vision with a mixture of sensors such as LiDARs, Cameras, and GPS to develop birds-eye-view semantically labeled and RGB-colored high-definition maps for autonomous vehicles • Optimized our HD map pipelines by replacing Python loops with NumPy array programming• Benchmarked image classifiers in Tensorflow and OpenVino (memory, disk-space, asynchronous and synchronous inference speeds, batched vs unbatched inference, ...) on Ford's on-prem clusters

Led a team of undergraduate researchers focused on evaluating and improving the consistency of CNN-Object Detectors. We worked on developing novel consistency metrics and applying new methods for improving consistency, such as anti-aliasing.Worked on low-power computer vision systems. I was the main contributor to an API in C++ which established server-client communication between an arbitrary number of clients. The purpose of this API was to send partitioned video streams over a network (TCP/IP) to a number of workers running 2D object detection.Sourced image data for the Low Power Image Recognition Challenge, 2018. Required image validation and labeling.General study in neural networks, explainable ai, dataset bias, and object detection.

Led the development of computer vision technology to detect and classify food and beverages in user-submitted photos for a technologically advanced dietary assessment application.

Helped close the generalization gap from synthetic to real data through the application of generative adversarial networks.