Led a team in the development of 0-1 MVPs for various applications, including an AI training platform, image generation service, edge AI camera, and AI surveillance system, successfully bringing innovative concepts to market.• Designed and developed an AI platform for computer vision training and inference, featuring an intuitive web UI for model training and deployment. (Flask, PyTorch, Docker, React.js)• Modularized core features, achieving an 80% reduction in compiling time and increasing system extensibility, allowing for quicker adaptation to changing requirements.• Implemented plugin architectures for data processing and deep learning models, fostering team collaboration and reducing feature development time by 40%.• Implemented testing and CI/CD pipelines, cutting manual actions by 50%, and identifying errors in advance (Gitlab runner, pytest)• Successfully migrated the AI platform to Azure, transforming it into a scalable system capable of handling increased workloads and user demands.• Improved deep learning model training time by over 50% through the implementation of advanced techniques such as parallelizing data pipelines and optimizing deep learning framework configurations, resulting in faster model iteration and deployment.• Enhanced inference response speed by 70% through optimizations in data format, service scaling, and the utilization of different data transmission protocols such as TCP & shared memory, as well as the partial migration of system components to C++, leading to improved real-time performance.• Designed and constructed an AI surveillance system with NVR-like functionality, integrating AI solutions and improving system efficiency by 3x, enabling the support of three 4k cameras simultaneously, leveraging FFmpeg and redesigning recording and frame processing workflows.