I built DeepMediScan, my final-year project at The Islamia University of Bahawalpur, to classify 22 skin diseases. I trained and benchmarked four deep-learning architectures and achieved a best test AUC of 0.9656.
For SkinCare AI, I built a FastAPI backend serving an ensemble of four trained models and developed a web front end with authentication, chat history, microphone input, and image upload or camera capture. I also integrated a Groq-hosted LLM as a conversational layer over the model predictions.
My work includes careful model evaluation, leakage-safe data splitting, and handling class imbalance and limited GPU memory. I also bring rubric-based quality evaluation experience, assessing model outputs against defined criteria and documenting findings.

