Shadab Haque
928-***-**** ***************@*****.*** linkedin.com/in/shadab-haque Guwahati, India Education
Virginia Polytechnic and State University Virginia, USA Master of Science in Computer Engineering Aug. 2022 – Aug. 2024 Jorhat Engineering College Jorhat, India
Bachelors in Technology in Electrical Engineering Aug. 2018 – June 2022 Technical Skills
Languages: Python, C/C++, SQL, R, Bash, Java, MATLAB/Simulink, Javascript Frameworks: PyTorch, Keras, Tensorflow, Hugging face, OpenCV, OpenAI API, PySpark, Scikit-learn, Scipy Tools & Technologies: Git, Docker, VS Code, PyCharm, Spyder, Jupyter, Numpy, Pandas, Matplotlib, Linux ML Algorithms: LLMs, CNNs, Transformers, PCA, Gaussian Processes, EM, VAEs, VITs, CLIP, XGBoost, YOLOv8 Experience
Jr. Computer Vision/Deep Learning Scientist (Client: Norfolk Southern) Sept. 2024 – Sept. 2026 Pyramid Consulting Remote, USA
• Collaborated with Business Stakeholders to enhance and improve automated AI railway defect detection
• Developed a novel augmentation method for data generation of mismatched rail-ends for a Line Transformer model
• Developed, implemented and optimized 8 deep learning models for unique railway track use cases
• Wrote clean, readable code using data structures and objected oriented programming and deployed them to production
• Implemented and optimized models for GPU based training and inference, including parallel execution on GPU clusters
• Utilized Active learning, from experimentation, edge cases and prototyping through deployment and monitoring Graduate Research Assistant May 2023 – Aug. 2024
Virginia Tech Virginia, USA
• Leveraged multimodal transformer models such as Grounding DINO and Segment Anything for plant stress detection
• Trained GroundingDINO for plant stress detection reducing false positives by 30% and increasing box mAP by 15%
• Utilized PointNet, PointNet++, PVCNN for the purpose of 3D plant stress detection and increasing F1 score by 15%
• Preprocessed point clouds using centering, normalization and fixed size sampling to increase 3D box mAP by 30% Computer Vision Intern Jan 2024 – Apr. 2024
Norfolk Southern Georgia, USA
• Collaborated with the Mechanical division within an agile framework to enhance digital train inspection
• Utilized a YOLOv8 keypoint model for knuckle face wear in locomotives, achieving a pose mAP 0.995
• Used negative sampling and synthetic image generation to account for edge cases making the model robust to outliers
• Achieved a 0.82 F1 score with a fine tuned EfficientNet-b0 to classify brake wheel chain paint for applied handbrake
• Post processed the results from the classifier and aggregated 2 camera views to further reduce false positives by 50% Publications
• Shadab Haque, Kanad Naleshwarkar, Gayatri Bhatambarekar, Zeel Desai, Aishwarya Kumaran, Adithya Harish Srinivisan (2023). Summarization and Classification of ETDs using LLMs. CS5604: Information Retrieval
• Silwal, A, Xhang, X. M., Hadlock, T., Neice, J., Haque, S., Kaundanya, A., Lu, C., Vinatzer, B.A., Kantor, G.,
& Li, S. Towards an AI-Driven Cyber Physical System for Closed-Loop Control of Plant Diseases. The Association of the Advancement of Artificial Intelligence
Projects
Data Valuation for Symbolic Regression Python, PyTorch, Numpy, Pandas Sept. 2023 – Dec. 2023
• Employed Transformer and MLP models on a dataset of 5K training, 1K validation and 1K test instances
• Demonstrated key data value in model accuracy; notable decline observed upon removal of high value data points
• Achieved and F-1 score of 0.331 in noisy label detection, effectively identifying and mitigating detrimental data
• Utilized SNIP for cross-modal property prediction, notably enhancing F-1 score for mathematical operator detection