This work presents a novel system that connects dynamic user profiles and adaptive feedback with language models, thereby rectifying the limitations of the usual recommendation systems currently deployed. The design will incorporate Retrieval-Augmented Generation (RAG) that customizes the retrieval step so that the content available in the domain is of an academic nature and that the answers are always straightforward and will suit new as well as seasoned researchers. In contrast to normal two-module splicing, our design pulls user modelling and retrieval layers nearer together, learning to automatically update profile information in real time. Along with that, we also established feedback loops, which enhance user embeddings based on the quality of the interaction experience and the degree of comprehension of the topics by the user. The framework applies constraints based on the particular academic fields, the extent and strength of the knowledge that people possess, which is not generic to typical generic recommendation algorithms. We tested the model on numerous experiments with known data sets and compared it to three alternative models. According to the researchers, the model increased the accuracy of recommendations, the quality of personalization, and user satisfaction. Furthermore, ablation studies have made it evident that failure to follow the proper design of the adaptive feedback loop or to properly lay out the fusion modules results in significantly worse performance. It proposes a versatile and intelligent answer to the suggestion of academic materials, a combination of technological and practical advantages of personal study.