Hasaan Hamid

PROFESSIONAL PROFILE Computer Science undergraduate at NUST (CGPA 3.5) with experience building end-to-end, production-grade AI systems across NLP, computer vision, mass spectrometry–based spectral analysis, and time-series forecasting. Strong ownership of the full ML lifecycle, from automated data acquisition (Selenium, BeautifulSoup) and large-scale preprocessing to transformer/CNN model development and optimization. Hands-on experience with RAG pipelines, PEFT/LoRA, quantization, knowledge distillation, and federated learning. Proven ability to deploy and scale ML services using FastAPI, Docker, Kubernetes, and cloud-native microservices with low-latency constraints. EDUCATION BS Computer Science SEECS , Islamabad , 3.54 (2026) INTERNSHIP EXPERIENCE truID 30-Jun-2025 - 22-Aug-2025 Engineered production-grade computer vision and NLP systems for identity verification, owning the pipeline from data acquisition and automated scraping (Selenium, BeautifulSoup) to model training, optimization, and deployment. Developed and benchmarked state- of-the-art CNN and Transformer architectures (Xception, EfficientNet, ConvNeXt, ViT) for card liveness detection and replay-attack prevention under real-world constraints. Built and fine-tuned custom M2M100-based transliteration models, applying PEFT/LoRA, knowledge distillation, and quantization to achieve low-latency inference. Optimized models for mobile and edge deployment using TensorFlow Lite, ensuring robustness, scalability, and sub-100ms performance in production environments. Machine Vision and Intelligent Systems Lab 01-Jun-2025 - 20-Aug-2025 Engineered continual learning pipelines using Avalanche to support incremental, non-stationary training regimes on high-dimensional biomedical and spectral datasets. Designed unified, cross-species architectures to replace legacy single-species models, emphasizing domain generalization, distribution shift robustness, and zero-/few-shot transfer. Implemented custom hybrid CNN– Transformer backbones and bespoke training loops in PyTorch, optimizing representation learning across heterogeneous modalities. Integrated model interpretability and calibrated uncertainty estimation, leveraging Grad-CAM, SHAP, Bayesian layers, and posterior predictive analysis to ensure reliable and explainable inference in safety-critical settings. FINAL YEAR PROJECT Pathoshield PATHOSHIELD is an AI-driven antimicrobial stewardship and surveillance system that predicts antimicrobial resistance (AMR) from MALDI-TOF mass spectrometry data. It replaces traditional rule-based diagnostics with deep learning models—both species-specific and cross-species—that generalize across hospitals, bacterial species, and evolving resistance patterns. The system uses hybrid CNN and CNN–Transformer architectures in PyTorch, employs continual learning to handle real-world non-stationarity, and integrates explainability (Grad-CAM, SHAP) and Bayesian uncertainty estimation for interpretable, trustworthy predictions. PATHOSHIELD enables real-time AMR surveillance, uncertainty-aware decision support, and electronic prescription integration, making it a scalable, deployable tool for modern microbiology labs and public health systems. TECHNICAL EXPERTISE

Computer Science undergraduate at NUST (CGPA 3.5) with experience building end-to-end, production-grade AI systems across NLP, computer vision, mass spectrometry–based spectral analysis, and time-series forecasting. Strong ownership of the full ML lifecycle, from automated data acquisition (Selenium, BeautifulSoup) and large-scale preprocessing to transformer/CNN model development and optimization. Hands-on experience with RAG pipelines, PEFT/LoRA, quantization, knowledge distillation, and federated learning. Proven ability to deploy and scale ML services using FastAPI, Docker, Kubernetes, and cloud-native microservices with low-latency constraints.