The increasing reliance on tele-therapy for mental health support highlights the need for advanced methodologies to improve diagnostic precision and patient outcomes. This study explores the transformative potential of transfer learning in natural language processing (NLP) to enhance the detection of mental health conditions during tele-therapy sessions. Leveraging a dataset sourced from mental health-related subreddits, which includes conversations mapped to five target categories (Stress, Depression, Bipolar Disorder, Personality Disorder, and Anxiety), we fine-tuned a pre-trained BERT model for multi-class classification. Our study's results highlight significant performance enhancements achieved through the implementation of transformer-based models. The proposed framework achieved an accuracy of 83%, with macro average precision, recall, and F1-score values of 0.84, 0.83, and 0.83, respectively. Class-specific analysis further underscores the model's robustness, with precision ranging from 0.75 to 0.92 and recall values exceeding 0.80 for most categories. These outcomes significantly outperform traditional machine learning models such as Random Forest (accuracy: 72.65%) and Support Vector Machines (accuracy: 69.71%), demonstrating the superior capacity of BERT to capture complex linguistic patterns and semantic nuances in patient interactions. This research underscores the transformative role of transfer learning in NLP applications for tele-therapy, offering a scalable and precise solution for mental health assessment and paving the way for personalized, AI-driven interventions.