Muhammad Athar

PROFESSIONAL PROFILE With a passion for mathematics and computer science, I am determined to bring about a positive change in the world through research and development by making things easier and more convenient for the general public. I am also interested in sharing my knowledge with anyone who is interested and believe that knowledge sharing is one of the key things that have led to the exponential growth of technology. EDUCATION Bachelors of Engineering in Software Engineering SEECS , Islamabad , CGPA: 3.9/4 (2026) INTERNSHIP EXPERIENCE TUKL-NUST R&D Center 03-Jun-2024 - 06-Sep-2024 Designed and Implemented a CNN + Transformer-based Deep Learning Architecture in PyTorch for the classification of EEG Signals into normal and abnormal which achieves State-of-the-Art performance on NMT Dataset collected in Military Hospital Rawalpindi. RheinMain University of Applied Sciences, Wiesbaden, Germany 10-Jun-2025 - 06-Sep-2025 Performed independent research on weather forecasting and devised a multi-modal flood forecasting framework which utilizes ERA5 weather data and global river discharge maps, trained on a Sentinal-1 satellite imagery based flood dataset. FINAL YEAR PROJECT EEGWriter: Multimodal Deep Learning Framework for Automated EEG Diagnostic Report Generation Electroencephalography (EEG) is a critical diagnostic tool in neurology, widely used for identifying abnormalities such as epilepsy, seizures, and other brain disorders. Interpreting EEG signals, however, requires expert neurologists and is a time-consuming process. This project aims to bridge the gap between raw EEG data and automated clinical interpretation using large language models (LLMs). EEGWriter proposes the development of a multi-modal architecture that can generate structured EEG diagnostic reports from time-series EEG signals, patient metadata (e.g., age, gender, symptoms), and technician's notes. The generated reports follow a medical report format, including Factual Report and Impression sections. The system consists of deep learning models that extract temporal and spatial features from raw EEG data using deep learning (e.g., CNN, Transformer models), and an LLM (e.g., Qwen-3 8B) that produces natural language text conditioned on these features. The final model will aim to simulate how a neurologist writes EEG interpretations, potentially supporting telemedicine platforms, automated triage systems, and AI-assisted diagnostics. TECHNICAL EXPERTISE C++ Honed my C++ skills through regular competitive programming on platforms like CodeForces and LeetCode. Python Extensively used Python throughout many of my development as well as research-based projects. Git Collaborated with many individuals via GitHub by utilizing Git. https://github.com/AtharRizwan

With a passion for mathematics and computer science, I am determined to bring about a positive change in the world through research and development by making things easier and more convenient for the general public. I am also interested in sharing my knowledge with anyone who is interested and believe that knowledge sharing is one of the key things that have led to the exponential growth of technology.