Abhishek Tarun

• Designed deep learning models to predict user presence up to 5 days in advance to reduce failed last-mile deliveries. • Achieved 9% higher precision over baseline by hyperparameter tuning using PyTorch. • Reduced model training time by ~30% by simplifying architecture, enabling faster iteration and deployment cycles. • Reduced overfitting in existing models by applying early stopping and dropout.

• Documented 300+ hardware assets in SQL and Excel for accurate inventory records. • Created relational mappings between workstations, devices, and users for data accessibility and consistency. • Mapped office network topology using Docusnap, connecting switches, servers, and Ethernet ports. • Standardized PC setup and documentation processes, enabling data-driven planning during hardware upgrades.