Depression, a profound mental health disorder, affects a vast number of people globally. Timely recognition and intervention are crucial to averting severe consequences such as suicide. Nevertheless, identifying depression proves difficult due to the subjective nature of symptoms, and societal stigma surrounding mental illness often discourages individuals from seeking professional support. Hence, this study aims to leverage state-of-the-art deep learning techniques to identify signs of depression in texts found on social media platforms. The study utilized word embedding techniques, such as FastText and Elmo to represent the text data in a vector format. This study employed various techniques such as SentiBERT, hybrid Bi-LSTM, and CNN + LSTM, to classify social media posts into non-depression, moderate and severe categories. To train the models, comments obtained from the social media platforms have been used. The results of this study demonstrate that the developed models can detect depression with 90% accuracy. The study suggests that the use of deep learning techniques can be a useful tool in detecting depression early and potentially preventing its development.