Muhammad Arbaz

• Fine-tuned a DenseNet-based face anti-spoofing model using Python, C++ and TensorFlow with 98% accuracy, enhancing biometric security in automotive infotainment systems. • Evaluated model expressions with ANTLR in Python to optimize decision boundaries reducing error rates by 20%. • Delivered a real-time demo framework with pixel-wise depth maps to highlight facial biometric capabilities.

• Presented thesis titled ”A Unified End-to-End Multi-Task Model for Face Recognition and Anti-Spoofing”. • Optimized a multitask model for resource constrained devices, achieving faster run times than single-task models. • Achieved a Half Total Error Rate of 4%, reducing error by 30% compared to previous state-of-the-art methods.

• Built a Python and Streamlit-based annotation tool for old historical documents, boosting process efficiency.

• Delivered machine learning-driven MVP for NFT price prediction using Python, securing $15K investor funding. • Designed a recommendation system to classify and recommend NFTs by style, boosting user engagement by 10%. • Troubleshooted and debugged machine learning models in TensorFlow, and PyTorch for optimal performance. • Trained MobileFaceNet for facial recognition and liveness detection for a digital identity platform. • Integrated CI/CD pipelines to automate the deployment of machine learning models.

• Refined ML ensemble models (random forests, XGBoost) and LSTMs to achieve a mean absolute error (MAE) of 5% for precise financial forecasting and analysis. • Created an FAQ chatbot and automated hospital claim resolution to improve operational efficiency. • Executed all stages of model development from data preprocessing to training and production deployment.