Depression is a rapidly expanding mental health issue that affects people psychologically, physically, and emotionally. This report outlines our work for DAIC (Distress Analysis Interview Corpus) as our third-year BTech project, addressing the rising need for automated depression detection using machine learning. Historically, identifying depression involved thorough clinical interviews, during which psychologists analyzed the subject’s responses to assess their mental well-being. In our model, we have constructed a custom convolutional neural network (CNN) on three different formats: text, video, and audio to predict the mental health status of the patient where the accuracy for the models are 86%, 85% and 90% respectively. We have implemented a deep learning architecture where each modality is assigned specific weights to collectively generate an output. This fusion approach addresses the following issues- handling noise in one of the modalities, regulating the impact of a specific modality on the overall result.