Emaan Arshad

I’m an AI researcher-in-training focused on applied machine learning and Explainable AI, driven by one recurring question: how do we make AI models transparent and reliable enough to be trusted in high-stakes, real-world decisions?

In my projects, I turn interpretability into audit-ready practice. For credit card fraud detection, I integrated SHAP into explainable anomaly detection under severe class imbalance, using SQL-driven pipelines across 280,000+ transactions; I improved recall by 17% while holding 82%+ precision and flagging 230+ anomalies, and I deployed a live demo on Hugging Face.

I also build end-to-end clinical and data-centric workflows. My diabetes risk pipeline trained on 253,680 patient records and implemented a full MLOps workflow (data versioning, pipeline automation, reproducible experiment tracking), comparing local and global interpretability using SHAP and LIME.

Beyond tabular ML, I work across computer vision and embedded systems—because trust must hold in every context. I developed a ResNet-based image aesthetics classifier with manual validation and correction of 1,200+ mislabelled records across an 18,000+ image dataset (sub-300ms real-time inference), and I designed a Smart Bin IoT stack using Arduino edge nodes, LoRaWAN, MQTT, and PostgreSQL, benchmarking 6 wireless protocols and implementing GDPR-compliant data handling. I’m now deepening my research direction through graduate study in artificial intelligence, with particular interest in explainable ML and uncertainty-aware, trustworthy models.