At COMSATS University Islamabad, I’m researching whether combining global information with local neighbourhood aggregation can make graph neural networks more robust to missing or spurious edges. I’m preparing baseline experiments on Cora in PyTorch.
As a Research Intern at the IT and Software Development Lab, ASCEND Tower, Islamabad, I built an end-to-end pipeline for ASCEND EchoLV using the CAMUS dataset. It combined U-Net segmentation, mask-derived measurements and an ejection fraction classifier; I also integrated the frozen pipeline into a Gradio research application.
At NeuroFive Solutions, I built and deployed a salon booking REST API with Node.js, Express and PostgreSQL. I added authentication, validation, tests, logging, cloud file storage and health monitoring.
My projects include a live face recognition attendance system, which reached 90% recognition accuracy on self-collected classmates’ faces, and ongoing work on a DDPM in PyTorch. I’m also comparing traditional machine learning classifiers with 1D CNNs for ECG arrhythmia classification.

