This paper introduces an advanced AI-driven assistive system designed to enhance accessibility for individuals with visual impairments by integrating real-time image detection, direct convolutional text-to-speech (DCTTS), and a large language model (LLM)-based decision-making module. The system leverages cloud-based architecture to ensure scalability and device independence, allowing seamless interaction across various platforms. By utilizing YOLOv8 and SSD for object detection, it identifies obstacles, traffic signals, and environmental landmarks, generating corresponding context-aware audio feedback through DCTTS. Additionally, the LLM-driven decision-making module dynamically interprets the user’s surroundings, providing intelligent navigation assistance, hazard warnings, and adaptive responses tailored to real world scenarios. To evaluate system performance, we analyze detection accuracy, speech latency, and user interaction efficiency, demonstrating a significant improvement in accessibility for visually impaired individuals. Compared to traditional rule based approaches, the proposed system offers greater adaptability, real-time responsiveness, and enhanced personalization. Future enhancements will focus on on-device inference for reduced latency, integration of vision-language models (VLMs), and personalized user feedback mechanisms to further refine assistive capabilities. This study highlights the potential of AI-powered accessibility solutions in fostering greater independence and safety for individuals with visual impairments.