Sami Sawani

Master’s Thesis On “HW-Aware Multi-Task Learning (MTL) for Computer Vision”. Derived multiple new HW-aware losses to optimize for latency on edge devices (Nvidia Jetson Orin and Qualcomm Gen2). Achieved more than 46% latency improvements on ONNX Runtime and Tensor RT. Built a custom edge friendly computation graph to leverage parameters sharing. Achieved up to 7% better accuracy on dense prediction tasks with 25% more parameters reduction compared to SOTA.

Proposed and implemented algorithmic optimization to better predict battery metrics like state of charge, and state of health. Achieved notable qualitative improvements for battery states predictions under varying operating conditions. Improved ‘voltage/ current’ predictions by up to 5%. Incorporated a newer dynamic hysteresis effect tracking that supported operating conditions not handled before.

Cofounded a new data analytics department. Established a proper framework and a cloud solution for aggregating, cleansing, and transforming Big Data from various Mobile Network Operators in Asia and Africa. Smarter campaigns-targeting with quantifiable cost savings.