The diagnosis of mental illnesses and their subcategories has long been subject to the clinician’s assessment and lacks an objective measurement criteria, leading to misdiagnosis and poor patient acceptance rates of their illness. Problem Statement - For patients with Bipolar Disorder, the anosognosia rate stands at 40%, thus they chronically refuse having an illness and avoid taking medication. Also, family members and those around these individuals are unable to objectively identify the impact of the illness on the patient, and this leads to poor social relationships and stigmatization of BD patients. Moreover, the functioning of BD patients is not efficient, as they do not have a proper way of tracking their illness. The intervention designed is a device prototype along with a mobile app aimed at using the non-invasive biomarkers of speech, heart rate, sleep and physical activity to objectively classify the affective states of Bipolar Disorder, and consequently communicate results to patients to provide guidance on the right course of action using evidence-based Cognitive Behavioral Therapy. The research was conducted in two parts. The first part was the assembly of components for the wearable electronic device that would monitor the symptoms and detect the phase. The main components included an ESP32 microcontroller, MPU6050, INMP441, MAX30100, and power supplies. The database included audio samples of a total length of 1 h for each affective state to extract audio features. Another database was also created with heart rate measurements for a total of 500 individuals in different BD affective states. A simple accelerometric device was built to collect sleep and physical activity data. Machine learning models were then trained in classifying the states based on the labeled data values. Finally, a random forest classifier was used to compile all the parameters and give the final prediction. The second part of the research was the development of a user interface based on Cognitive Behavioral Therapy to assist the patient. The wearable device for BD phase prediction developed by integrating the INMP 441, MAX30100, and MPU6050 sensors with the ESP32 WiFi microcontroller reliably collected the real-time sensor data and transmitted it to the models for predicting the BD phase. The Random Forest classifiers accurately classified heart rate and physical activity data into the five bipolar disorder phases: mania, hypomania, depressive, euthymic, and mixed. Similarly, the CNN model demonstrated strong performance in classifying speech data into the same phases. The device consistently provided real-time feedback on the user’s BD phase across multiple trials.