Yamuna Sugumar

Computer Vision Engineer - General Motors - Canada

(2025-04)

Developed computer vision software using Python and C++ to process visual data for automotive perception applications, building reliable production-oriented components that supported real-time engineering and advanced vehicle technology initiatives.
Implemented AI-driven vision workflows using deep-learning techniques for visual data analysis and perception tasks, converting image inputs into actionable model outputs for evaluation and iterative improvement by computer vision engineering teams.
Optimized image-processing pipelines for latency-sensitive computer vision workloads, improving computational efficiency and supporting the responsiveness required by real-time automotive perception systems.
Evaluated computer vision model performance through systematic testing, debugging and failure analysis, identifying challenging visual scenarios and translating findings into improvements in model robustness and overall system reliability.
Integrated vision algorithms into software workflows using C++, Python and Git, supporting collaborative development and iterative deployment across cross-functional engineering teams during 16+ months of automotive software development.
Contributed to AI and computer vision initiatives by translating image-processing and machine-learning concepts into scalable software components, supporting reliable perception capabilities for automotive applications and production-oriented engineering workflows.

Software Engineer (Computer Vision) - Yantravision Software Pvt. Ltd. - India

(2021-05 - 2023-03)

Designed OpenCV-based image-registration algorithms for automated industrial inspection, improving registration accuracy to 98% and enabling consistent image alignment and print quality across production workflows.
Built end-to-end computer vision tools using Python and C++ for image alignment, defect detection, localization and image augmentation, automating inspection processes and improving the reliability of print-quality analysis.
Engineered high-efficiency CUDA kernels for computationally intensive image-processing operations, reducing processing latency from 50+ ms to 19 ms and enabling the inspection system to meet real-time performance requirements.
Profiled GPU workloads using NVIDIA Nsight and NVIDIA Visual Profiler, identifying processing bottlenecks and optimizing execution paths to improve performance across high-throughput computer vision pipelines.
Streamlined automated inspection workflows by optimizing image-processing stages and execution efficiency, increasing inspection speed by 45% while improving system reliability and minimizing operational downtime.
Delivered scalable computer vision solutions covering image registration, defect detection and localization, translating vision algorithms into production-ready inspection capabilities for industrial manufacturing applications.

About