Early diagnosis of developmental disorders is crucial for timely intervention, yet traditional assessment methods rely heavily on subjective evaluations, which can be time-consuming and prone to bias. This study aims to develop an AI-powered system capable of differentiating between scribbles made by special children and those made by typical children. The system integrates Convolutional Neural Networks (CNNs), Autoencoders, and Random Forest classifiers to analyze key features such as stroke pressure, line consistency, spatial distribution, and pattern complexity. A dataset of scribbles is collected through both digital and paper-based methods, followed by preprocessing techniques including image resizing, normalization, noise reduction, and data augmentation to enhance model performance. The AI-driven approach offers an objective, scalable, and efficient tool for early diagnosis in educational and medical settings. Results indicate high accuracy in classification, demonstrating the potential of this system as a diagnostic aid for detecting developmental challenges in children. Future work includes expanding the dataset, integrating real-time handwriting dynamics, and developing a user-friendly application for broader accessibility.