Syed Ibrahim Shakir

- Distilled vision foundation models for knowledge transfer and model compression, boosting student classification accuracy by 21% - Implemented and evaluated face detection models (RetinaNet, CenterNet, etc) for downstream news media analysis, improving performance by 10-30% across tasks - Explored federated learning as an alternate paradigm for decentralized training in privacy-sensitive applications - Tools: PyTorch, Lightning, Python, Kubernetes, Docker, MLOps, Pandas & Jupyter

- Conducted thorough study & evaluation of state-of-the-art object tracking algorithms (e.g., Deep SORT, SiamMask) against real-world data (TAO dataset), identifying model-specific performance limitations in diverse real-world scenarios - Implemented and benchmarked multiple object tracking algorithms, analyzing performance in detection, localization & association using HOTA metrics (achieved 13.3 cumulative score improvement) - Tools: Python, PyTorch, OpenCV, Pandas, NumPy, sklearn, & Jupyter

• Extended bachelor’s thesis work on RAES (Rehabiliation Assistance Exo-Skeleton) • Designed an EMG sensor using FET-based differential amplifier & bandpass filtering • Provided IoT solutions for smart home automation • Tools: Python, Embedded C/C++, OpenCV, Arduino & Raspberry Pi

• Applied computer vision techniques for stereo depth estimation and real-time object localization in robot navigation • Analyzed human & animal gait patterns to balance robots with shifting center of gravity • Designed a control system for synchronous joint actuation of quadruped robot • Tools: Python, Embedded C/C++, Rotary encoders, IMUs, OpenCV & Stereo Vision

• Used real-time computer vision (OpenCV) to verify ball throws through target hoops on a Raspberry Pi 3, achieving sub-250ms latency per frame • Built a vision-based color recognition pipeline, replacing legacy laser sensors to trigger robotic task state-machines during the match sequence • Programmed a line-following robot using a PD control loop, reducing deviation from track & enhancing stability in motion of the autonomous robot • Tools: Arduino, ATmega, Raspberry Pi, Embedded C/C++, Python, Matlab

Thesis: Efficacy of Foundation-Model-Based Distillation for Image Classification

Rehabilitation Assistance Exo-Skeleton