Machine Learning Engineer, Computer Vision2021 - PRESENTBusPatrol
Technologies: PyTorch, Python, Docker, OpenCV, SciPy, Machine Learning, Artificial Intelligence (AI), Machine Learning Operations (MLOps), Amazon Web Services (AWS), You Only Look Once (YOLO), NumPy, Deep Learning, Computer Vision, Image Processing, Scikit-image, Scikit-learn, TensorFlow, Data Science, Convolutional Neural Networks, Amazon SageMaker, Infrastructure as Code (IaC), Pandas, 3D Image Processing, Object Detection
- Contributed to several computer vision projects using state-of-the-art CNN and vision transformers to solve a boundary detection problem with over 98% accuracy and optimized the neural networks to double the speed gain on CPU inference.
- Led the deployment of a machine learning (ML) lifecycle management infrastructure on AWS, increasing efficiency and reproducibility of ML workflows internally and easing collaboration with the DevOps team.
- Addressed a business-critical problem, which was first thought to need AI, through simple data aggregation and analysis. The surfaced insights brought a simple way to solve the issue, saving the company significant time and cost.
- Deployed production-grade ML Inference Infrastructure to AWS as microservices using AWS CDK and AWS SageMaker Endpoints.
Machine Learning Engineer, Computer Vision2020 - 2021Faimdata
Technologies: TensorRT, NVIDIA Jetson, OpenCV, PyTorch, Python, C++, CUDA, CMake, Google Cloud Platform (GCP), Node.js, Docker, DeepStream SDK, Artificial Intelligence (AI), Swift, You Only Look Once (YOLO), NumPy, Deep Learning, Computer Vision, Image Processing, Machine Learning, SciPy, Scikit-image, Scikit-learn, TensorFlow, Data Science, Convolutional Neural Networks, Pandas, 3D Image Processing, Object Detection
- Developed the company's first optical character recognition algorithms and pipeline in Python, C/C++, PyTorch, TensorRT, and DeepStream, succeeding with 95% accuracy in the first pilot project, leading to the company acquiring the client's business.
- Applied transfer learning from the first deployments to a new use-case, reproducing similar performance levels, succeeding in a second pilot project that led to the acquisition of a second client's business.
- Reduced iteration time by 25% after establishing workflow best practices in ML deployment iterations using Docker, Google Cloud Platform, Git, Data Version Control, and model testing.
- Pruned deep neural networks (DNNs) to 60% faster runtime, allowing for the use of more affordable hardware and reducing hardware expenses by 15%.
- Hired and trained a new team member and transferred knowledge to help grow and scale the company's computer vision team.