CV

Visiting Research Scientist

STARTS, Inria - French National Institute for Research in Digital Science and Technology

Valbonne, France

• Conducted collaborative research on computationally efficient ML methods for personalized human behavior and expression analysis.
• Investigated test-time multimodal domain adaptation approaches, eliminating the need for original training data while improving model robustness and efficiency.
• Performed a technology watch on emerging AI/ML research relevant to efficient personalized modeling.

Computer Vision Researcher

LIVIA & ILLS Labs - École de technologie supérieure (ÉTS)

Montreal, Canada

• Develop deep learning models for visual behavior and expression analysis using Python and PyTorch.
• Design personalized and domain-adaptive computer vision methods to improve robustness across subjects, environments, and data conditions.
• Propose multimodal learning approaches integrating visual, physiological, and audio signals for robust prediction in real-world settings.
• Build end-to-end ML pipelines for data preprocessing, training, evaluation, ablation studies, and reproducible experimentation.
• Conduct large-scale experiments to evaluate accuracy, robustness, generalization, and cross-domain performance on noisy real-world datasets.
• Co-developed the BAH dataset for ambivalence and hesitancy recognition, including benchmarking and evaluation protocols.
• Collaborate with interdisciplinary researchers to refine model design, evaluation criteria, and experimental methodology.

Senior Software Engineer

Jin Technologies

Production software systems and data-driven applications

• Built scalable backend systems and RESTful APIs for production data-driven applications.
• Designed modular, maintainable software components with a focus on reliability, performance, and scalability.
• Collaborated with cross-functional teams to translate product requirements into production-ready systems.
• Applied software engineering best practices, including debugging, documentation, version control, and system optimization.

Software Developer - Video Content Retrieval

Bahria University Research Lab

Applied computer vision and multimedia retrieval

• Developed computer vision pipelines for Urdu text detection and recognition in low-quality video streams.
• Built indexing and retrieval systems for large multimedia datasets, enabling visual content search.
• Designed data preprocessing workflows for noisy real-world video data, including low-resolution frames and variable lighting conditions.
• Applied Python, OpenCV, and TensorFlow to build and evaluate applied image/video processing models.

Android Developer

Steprobotics

Mobile computer vision for solar irradiance estimation

• Developed computer vision algorithms for skyline and shading analysis to support solar irradiance estimation.
• Built the Step Solar Android application, integrating geospatial analysis, mobile data capture, and external APIs.
• Designed image-processing workflows for real-world outdoor visual data with variable lighting and scene conditions.
• Collaborated with engineering teams to integrate CV outputs into a user-facing mobile application.

Ph.D. in Systems Engineering

École de technologie supérieure (ÉTS)

Deep Learning, Multimodal AI, Domain Adaptation, Human Behavior Analysis

M.Sc. in Data Sciences

Bahria University

Data Sciences

B.Sc. in Computer Sciences

Air University

Computer Sciences

MuSACo: Multimodal Subject-Specific Selection and Adaptation for Expression Recognition with Co-Training

IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)

Subject-specific multimodal source selection and adaptation for robust expression recognition.

Personalized Feature Translation for Expression Recognition: An Efficient Source-Free Domain Adaptation Method

International Conference on Learning Representations (ICLR)

Efficient source-free domain adaptation method for personalized expression recognition.

BAH Dataset for Ambivalence/Hesitancy Recognition in Videos for Digital Behavioural Change

International Conference on Learning Representations (ICLR)

Co-developed a publicly available benchmark for ambivalence and hesitancy recognition in behavioral analysis.

Research Impact & Publications

Published in top-tier AI venues including ICLR, WACV, and CVPR Workshops.

• 300+ citations and h-index of 8 on Google Scholar.
• Co-developed the BAH dataset, a publicly available benchmark for ambivalence and hesitancy recognition in behavioral analysis.

FRQNT Award

Fonds de recherche du Québec - Nature et technologies

Personalized Deep Learning Models for Diverse Individuals, Quebec, Canada.

4-Year Ph.D. Scholarship

École de technologie supérieure (ÉTS)

Fully funded research position at ÉTS, Montreal.

Top 100 - Hello Tomorrow Global Summit

Hello Tomorrow

Recognized at Le Centquatre, Paris, France.