The integration of computer vision into real-world applications is often hindered by the high computational demands of deep learning models and the associated cost of hardware. This paper presents a novel, comprehensive solution for the real-time digitization of a physical chessboard, specifically designed for low-cost embedded platforms. We introduce a complete system that combines efficient classical computer vision algorithms for chessboard localization with a custom-designed, lightweight Convolutional Neural Network (CNN) for piece recognition. The core of our innovation lies in the use of a low-cost, open-source RISC-V platform, which is augmented with a custom, tightly coupled hardware accelerator to offload the most computationally intensive tasks. Our system demonstrates a chessboard localization accuracy of 99.5% and a piece recognition accuracy of 98.2%. Critically, it achieves an average processing latency of 150 milliseconds per frame, enabling true real-time operation on hardware that is an order of magnitude more affordable and power-efficient than conventional GPU-based systems. This work validates a crucial hardware-software co-design paradigm for deploying complex AI tasks on resource-constrained devices, paving the way for a new generation of affordable, intelligent edge applications.