Intelligent Tutoring Systems (ITS) and Augmented Reality (AR) have become greatly popular in current scenario, especially for helping students in mastering difficult subjects through a variety of different methods with the implementation of smart algorithms. There are many papers in the current literature that discuss the ITS architecture and the AR architecture independently; a few papers have even proposed designs for combining these systems, but the need for this article arises in order to suggest improvements that could theoretically increase the performance and overall robustness of the system for learning basic, complex, domain-specific and AR related concepts. This article discusses the existing ITS and AR systems and their flaws, followed by some potential benefits that can be achieved by combining ITS and AR effectively. We propose a novel architecture for improving the combined AR and ITS system scalable for supporting interaction for the diverse users and domain. The proposed system makes an effective use of three tier architecture, load sharing algorithms, data management techniques, multiple servers, marker-less AR, and modeling 3D object models on the fly, in order to make the system more effective, secure, reliant, and seamless for the users. For realizing 3D object modeling on the fly, the article presents an improved method by combining Level of Detail and Rasterization techniques in order to render in steps in accordance with the demand (i.e. processing up to adequate and sufficient level of details), which will help us use the architecture for small scale to large scale systems. Although 3D object modeling on the fly needs storage up to 33% more than the conventional geometrical structure of the mesh, the speed-up achieved can be as high as six times for coarse mesh and up to 1.46 times for fine mesh. At the core of the proposed system, is to make the ITS extendible to multiple domains of learning and education, and to reduce the response time and latency.