• Initiated deep-learning research in the detection of metastatic cancer in tissue samples. Investigated multiscale detection algorithms to identify cancerous tissue in high-resolution slide scans. • Developed state-of-the-art vision transformer networks to model changes in protein structure from multispectral observations of tissue and broad variations in tumor sizes that occur across different medical imaging datasets.• Enhanced the ability of vision-based transformers to generalize in biomedical segmentation, brain tumor detection, and medical analysis across a diverse set of datasets and histological variations.-

• Deployed a deep learning solution for smart parking that helped achieve a 600% reduction in traffic congestion and manage 35,000 weekly requests, employed with MS-SQL Server, C#, Scikit-Learn,Python, Pytorch, and FastAPI.• Engineered an attention-based encoder-decoder network for multiline optical license plate recognition, enhancing accuracy by 2.5% while achieving real-time processing capability at 40 images per second.• Achieved a 12% improvement in recognition accuracy through computer vision algorithms designed to normalize skewed object angles. Implemented contour approximation, binary search, and convex hull point reordering for precise detection and alignment of corners and angles based on object segmentation results. • Developed asynchronous backend features, implemented the singleton object pool design pattern, and incorporated multi-worker functionality to handle concurrent requests efficiently.• Managed and supervised a crowdsourcing team to build datasets and perform data preparation for deep learning with over 200,000 license plate images within project timelines, ensuring data accuracy and quality control.

• Administered an advanced Multimedia Analysis course for over 100 students, developed 5 weeks of targeted lab assignments, and orchestrated an impactful curriculum for an enhanced learning experience.

• Collaborated on a NeurIPS 2022-accepted research: Conference Living Lab (ConfLab). Introduced five statistical modeling metrics in Python for evaluating continuous pose data, resulting in a 20% improvement in data reliability.• Enhanced ground-truth pose data for 14 videos through the development of a YOLOv7 people detection model and a pose-matching algorithm, facilitating more precise motion tracking and automatic error correction.

• Instructed, mentored, and evaluated over 500 students in an advanced Algorithm Design class. Offering constructive feedback on assignments and exams, fostering a supportive educational environment.

• Coordinated with cross-department teams to build a P&L management application, streamlining mapping for over 3000 financial accounts and reducing reporting time by 45%.• Implemented core data transfer and hierarchical structuring features using Typescript, Node.js, npm, and React. Enhanced response performance for high data volumes by incorporating a lazy tabular data loading algorithm.• Accomplished a 50% user error rate reduction and resolved 15 critical design issues and performance bottlenecks through extensive software testing, debugging, code review, and use case testing conducted with UI/UX experts.