Harshil Gorasiya

- Developed web-based solutions by translating business requirements into user-focused interfaces and application workflows. - Contributed to web projects involving AI-enabled solutions, integrating functional requirements into practical user experiences. - Conducted user research, wireframing, prototyping, and iterative optimization to improve usability and consistency across web applications.

- Developed TensorFlow/Keras CNN with residual blocks for age and gender classification on 22,000+ images - Improved validation accuracy from 71% to 79% (+8 percentage points) through architecture search and class-balanced data augmentation - Reduced training time by 20% using cosine learning-rate scheduling and batch-size tuning with zero accuracy loss - Diagnosed and resolved data-leak issues across 10+ structured experiment runs, ensuring reproducibility and reliability

- Implemented YOLOv8n vehicle detection pipeline on pure CPU edge hardware (Intel Core i7), achieving 35,000+ frames/hour at 28–32 FPS - Achieved 2.4× higher throughput compared to YOLOv5s baseline on the same CPU-only infrastructure - Curated and annotated 3,000+ images across five vehicle classes; optimized IoU and NMS thresholds to reduce false positives in low-light conditions by 18% - Bundled complete inference and visualization pipeline into a single, deployment-ready Python application with performan