Nowadays, anxiety and stress have become increasingly prevalent among young people, particularly those transitioning from high school to college. This paper focused on predicting the stress level of the students in a specific age group (16–25) using multi-modal data. Additionally, we have explored the various causes of stress and their intensities. We examine the efficiency of several pre-trained AI models (LLM, Gemma, Llama, Mistral), and the machine learning model Support Vector Machine to develop student stress prediction. Also, this paper has proposed a multi-modal RAG AI model for generating a specific question-answering purpose. In that case, LLM uses context retrieval and its text generation knowledge to provide the answer to the user. Both offline and online questionnaires are used for collecting the dataset. This paper has provided result comparisons in terms of accuracy, precision, and recall and also provides BLEU and ROUGH score metrics for measuring the performance of AI models.