MINDWELL AN AI-POWERED MENTAL HEALTH COMPANION

International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 110 MINDWELL AN AI-POWERED MENTAL HEALTH COMPANION Dr. Usha KC 1 #1 , Sampreeth S Shetty 2 #2 , Shravya Ravindra 3 #3 , Thanisha K 4 #4 #1 Dr. Usha KC, Assistant Professor, Computer Science and Engineering DSATM, Bengaluru, India, [email protected] #2 Sampreeth S Shetty, Student, 2 nd year B.E, Computer Science and Engineering, DSATM, Bengaluru, India, [email protected] #3 Shravya Ravindra, Student, 2 nd year B.E, Computer Science and Engineering, DSATM, Bengaluru, India, [email protected] #4 Thanisha K, Student, 2 nd year B.E, Computer Science and Engineering, DSATM, Bengaluru, India, [email protected] ARTICLE INFO ABSTRACT ©2025 RS Publication Paper ID: IJASTR6927253B31105 Received: 2025-10-27 Published: 2025-11-26 DOI: https://dx.doi.org/ 10.5281/zenodo.1772 6758 Page No: 110-128 In recent years, mental health has emerged as a critical concern, especially among students and working professionals who face high levels of stress, anxiety, and emotional fatigue. While wearable devices like smartwatches can track physiological parameters such as heart rate and sleep, they often fail to interpret these signals in relation to a user’s emotional or psychological wellbeing. This project proposes an AIpowered Mental Wellbeing Monitoring System that intelligently analyzes both physiological and emotional data to detect stress and provide personalized support in real time. The system integrates heart rate data (from a smartwatch or simulated input) with natural language processing (NLP) techniques that evaluate the emotional tone of user text input. Using these combined insights, it computes a wellbeing score and offers instant recommendations such as breathing exercises, relaxation tips, or optional counselor connections. A Streamlet-based interactive dashboard visualizes live stress levels, emotional trends, and chatbot interactions, enabling users to monitor their mood patterns effortlessly. This uses VADER, a lexicon and rule-based algorithm, on your backend to analyse the sentiment and emotional tone of the user's text. Second, you leverage Neural Machine Translation (NMT), a deep learning model, via the Goslate library to translate chatbot responses. Finally, you use the browser's Web Speech API, which contains two "black box" AI models: Automatic Speech Recognition (ASR) to transcribe the user's voice to text, and Text-to-Speech (TTS) to generate spoken audio from the chatbot's replies. This approach bridges the gap between physical and emotional health tracking, promoting early stress detection and selfawareness. By leveraging artificial intelligence, the system aims to create a stigma-free, accessible, and personalized mental wellness companion that empowers individuals to take proactive steps toward maintaining a balanced and healthier mind. Future enhancements include integration with Google Fit or Apple Health APIs, voice-based emotion detection, and personalized meditation content for holistic wellbeing. Key words: Mental Wellbeing, Multi-modal Data Fusion, Sentiment Analysis, Heart Rate Variability (HRV), Electron, AI Chatbot Natural Language Processing (NLP), Speech Recognition (ASR),Machine Translation. International Journal of Advanced Scientific and Technical Research Available online on http://www.rspublication.com/ijst/index.html ISSN 2249-9954 Cite This Paper: THANISHA K, Dr. Usha K.C, Sampreeth S Shetty and Shravya Ravindra (2025). "MINDWELL AN AI-POWERED MENTAL HEALTH COMPANION". INTERNATIONAL JOURNAL OF ADVANCED SCIENTIFIC AND TECHNICAL RESEARCH (IJASTR), vol. 15, no. 6, 2025, pp. 110-128. DOI: https://dx.doi.org/10.5281/zenodo.17726758 International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 111 INTRODUCTION In an era where mental health challenges are increasingly prevalent yet often underdiagnosed, the integration of physiological and psychological monitoring offers a promising avenue for proactive wellbeing management. Traditional methods of stress assessment typically rely on subjective selfreporting, which can be inconsistent, or isolated physiological metrics, which lack emotional context. To address this gap, this paper presents a multi-modal mental wellbeing monitoring system that fuses realtime physiological data with natural language processing. By synergizing heart rate variability (HRV) data from wearable IoT devices with sentiment analysis derived from user interactions, the proposed system provides a holistic assessment of an individual's mental state. This unified approach not only detects stress with greater accuracy but also enables timely, personalized interventions through an AI-driven recommendation engine, thereby bridging the disconnect between physical health tracking and emotional care. The proposed system is implemented as a cross-platform desktop application using the Electron framework to ensure accessibility and user privacy. Its core architecture features a Flask backend that manages user data via an SQLite database and orchestrates the AI-driven analytics. This includes two primary components: 1) a physiological analysis module that processes time-series data from wearables (e.g., heart rate) and 2) a natural language processing (NLP) engine. This engine utilizes the VADER sentiment analysis algorithm to score the emotional valence of user text inputs, supplemented by the Web Speech API for voice-to-text transcription and Goslate for multilingual support. The primary objective of this study is to demonstrate the feasibility of this fused architecture in providing a real-time, correlated "Wellbeing Score," presented to the user through an intuitive Chart.js dashboard, thereby offering a more nuanced and actionable tool for personal mental health awareness. The system is presented as a proof-of-concept prototype, with its primary objective being to demonstrate the feasibility of fusing these specific data streams within a secure, self-contained desktop application. It is not intended to serve as a clinical diagnostic tool, nor does it replace professional medical advice or intervention. The analyses are based on consumer-grade sensor data and a general-purpose, pretrained sentiment model (VADER), which carry inherent limitations in precision compared to clinicalgrade equipment or domain-specific psychological models. Therefore, the system's aim is to enhance user self-awareness and provide a first-line mechanism for personal wellbeing tracking, rather than to offer a definitive medical or psychiatric diagnosis. LITERATURE SURVEY This review synthesizes existing research in three key domains relevant to the proposed project: 1) the use of physiological signals, particularly Heart Rate Variability (HRV), for stress detection; 2) the application of Natural Language Processing (NLP) for sentiment analysis in mental health; and 3) the established role of multi-modal systems and AI chatbots in wellbeing interventions. 1. Physiological Stress Detection via Heart Rate Variability (HRV) The link between psychological stress and the autonomic nervous system (ANS) is well-established in psychophysiological research. Stress activates the sympathetic nervous system (the "fight-or-flight" response), leading to a decrease in parasympathetic (the "rest-and-digest") activity. Heart Rate Variability (HRV)—the measure of variation in time between successive heartbeats—has emerged as a robust, noninvasive proxy for quantifying this ANS balance. International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 112 A significant body of literature confirms that acute psychological stress is reliably correlated with a reduction in HRV. Studies by Kim et al. (2018) and Castaldo et al. (2015) demonstrate that various timedomain (e.g., SDNN, RMSSD) and frequency-domain (e.g., the LF/HF ratio) metrics of HRV are significantly altered during stressed states. Specifically, reduced parasympathetic modulation, indicated by a lower RMSSD and high-frequency (HF) power, is a consistent marker of mental load and stress. This extensive validation supports the use of HRV, captured by consumer-grade wearables, as a reliable physiological input for a computational stress model. 2. Sentiment Analysis for Psychological State Assessment While physiological data reveals that a user is stressed, text-based sentiment analysis can provide context as to why. The field of Natural Language Processing (NLP) has been widely applied to assess mental wellbeing by analyzing user-generated content from social media, private journals, and chatbot interactions. Guntuku et al. (2017) demonstrated that language patterns on social media can predict depression and other mental health conditions. For real-time applications, lexicon-based models like VADER (Valence Aware Dictionary and sEntiment Reasoner) are particularly effective. As detailed by Hutto & Gilbert (2014), VADER is specifically attuned to the nuances of social media and informal text, incorporating rules to interpret capitalization, punctuation, and negations (e.g., "not good"). This makes it superior to traditional models for short, informal inputs, such as a chat message or a brief journal entry. Its use in mental health applications is validated as a computationally "lightweight" yet effective method for extracting emotional valence, a critical component of the user's psychological state. 3. Multi-modal Systems and AI Chatbots for Intervention Recognizing that neither physiological data nor sentiment analysis alone can capture the full complexity of human wellbeing, recent research has converged on multi-modal systems. A survey by García-Ceja et al. (2018) on multi-modal sensing for mental health monitoring concludes that fusing heterogeneous data streams—such as physiology, speech, and text—yields a more accurate and holistic assessment than any single modality. This project's "Fusion Logic" directly aligns with this advanced approach, correlating the physical stress (from HRV) with the emotional context (from VADER) to generate a comprehensive wellbeing score. Furthermore, the role of AI-driven conversational agents (chatbots) in delivering mental health interventions is rapidly expanding. As summarized by Abd-Alrazaq et al. (2020), mental health chatbots have proven effective in providing psychoeducation, delivering cognitive-behavioral therapy (CBT) exercises, and suggesting coping strategies (such as relaxation techniques). They offer an accessible, scalable, and non-stigmatizing first line of support. This project extends this paradigm by using the AI chatbot not only as an intervention tool but also as a data-gathering component, creating a continuous feedback loop where the user's state informs the chatbot's recommendations. Finally, the architectural choice of using Flask and Electron addresses a critical challenge in mental health technology: privacy. Much research highlights user concerns over sensitive health data being processed in the cloud. By packaging the Python backend and database into a local desktop application, the proposed system ensures that all personal data remains on the user's machine, a significant ethical and practical contribution to the field of personal health informatics International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 113 METHODOLOGY This study's methodology is centered on the design, development, and integration of a multi-modal system for real-time mental wellbeing assessment. The system is architected as a self-contained, crossplatform desktop application using the Electron framework to ensure all user data remains on the local machine, thereby addressing critical privacy concerns prevalent in cloud-based health applications. The application's architecture features a lightweight Flask web server packaged within the Electron container, running as a background process to manage all data processing, handle API requests from the frontend, and interface with the local database. For data persistence, a SQLite database is utilized, reinforcing the local-first, privacy-centric design by storing all user accounts and health data in a single file on the user's computer. User authentication is managed by the Flask backend using the Werkzeug Security library to perform secure one-way hashing of passwords. The system's operation relies on two concurrent data-processing pipelines. The first is a physiological pipeline designed to process time-series data from wearables, which begins with the user uploading a CSV file containing heart rate (HR) and Heart Rate Variability (HRV) metrics. The backend receives this file and employs the Pandas library to read the data into a DataFrame for essential wrangling, cleaning, and feature extraction, such as calculating aggregate stress indicators. This processed physiological data is then saved to the SQLite database via Flask-SQLAlchemy, linking it to the user's profile. International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 114 Running in parallel, the psychological data pipeline analyzes user text input from a journal or chatbot interface. The user can interact by typing or speaking; for voice input, the frontend JavaScript leverages the browser's built-in Web Speech API for speech-to-text transcription. This text string is then sent to a Flask server endpoint, where the core analysis is performed by VADER (vaderSentiment), a lexiconand rule-based sentiment model chosen for its high accuracy on short, informal text. To enhance accessibility, the system's English-based response is passed to the Goslate library for machine translation before being returned to the user, and the Web Speech API is used again for text-to-speech synthesis of the translated reply. The final stage of the methodology is the synthesis of these two disparate data streams. The system's core innovation lies in its fusion algorithm, which normalizes the outputs from both the physiological pipeline (stress score) and the psychological pipeline (sentiment score) to a standard range. A weighted-average algorithm then combines these normalized scores to calculate a single, unified "Wellbeing Score." This final score, along with historical data, is passed to the frontend, where the Chart.js library renders it on the user's dashboard in intuitive visualizations, such as line graphs for "Stress Over Time" or "Mood Trends." This allows the user to easily monitor their wellbeing and identify correlations between their activities, feelings, and physiological stat SYSTEM ANALYSIS AND DESIGN The design and analysis of the Mental Wellbeing Monitoring System were guided by a formal requirements analysis, which established the functional and non-functional goals of the project. The primary functional requirements identified were: 1) secure user authentication and session management; 2) concurrent ingestion of two distinct data types: physiological time-series data from CSV files and realtime user text/voice input; 3) AI-driven processing of these inputs, including sentiment analysis and language translation; 4) a core fusion logic to combine physiological and psychological metrics into a single, unified "Wellbeing Score"; and 5) an intuitive data visualization dashboard for user feedback. The non-functional requirements were paramount, mandating strict user privacy through a local-first data model, cross-platform compatibility (Windows, macOS, Linux), and a responsive, real-time user interface. Getty Images Based on these requirements, a client-side, multi-tiered architecture was designed, which was then packaged using the Electron framework. This architectural choice was critical as it directly fulfills the non-functional requirements for privacy and cross-platform compatibility. Electron acts as a native application wrapper around the webbased application (frontend) and a locally executed Python server (backend), ensuring that no sensitive user data ever leaves the user's machine. This design effectively creates a self-contained, three-tier system: a Presentation Tier (the UI), a Logic Tier (the Flask server), and a Data Tier (the local database). International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 115 The Presentation Tier is constructed using HTML, CSS, and JavaScript, forming the user-facing part of the application. This tier is responsible for rendering all visual components, including login forms, the chat interface, and data dashboards. It uses JavaScript's fetch API to send all user inputs (form data, text messages, file uploads) to the local Flask server and to receive processed data back. This tier also integrates two key browser APIs: the Web Speech API for both speech-to-text (STT) transcription of user voice commands and text-to-speech (TTS) synthesis of the chatbot's replies, and Chart.js for rendering the dynamic, interactive graphs of stress levels and mood trends on the user's profile. The Logic Tier operates as the application's "brain," running as a lightweight Flask web server in a background process managed by Electron. This Python backend exposes a series of API endpoints to handle all business logic. When a user logs in, this tier uses Werkzeug Security to check hashed passwords against the database. When data is received, it directs it to the appropriate processing pipeline. For text input, it calls the VADER library to perform sentiment analysis and the Goslate library to translate responses. For file uploads, it uses the Pandas library to read, clean, and analyze the physiological data from the CSV, extracting key features like mean heart rate or stress indicators. Finally, the Data Tier is implemented using SQLite, a serverless, self-contained database engine. This choice is integral to the privacy-first design, as the entire database, including user credentials and all historical health data, is stored in a single file on the user's local disk. The Flask-SQLAlchemy ORM (Object-Relational Mapper) is used by the Logic Tier to provide a simple and secure interface for all database operations, such as creating new users, saving processed sensor data, and retrieving historical data for the fusion algorithm. This design ensures that the application is fully functional offline and that the user retains complete control over their personal information. International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 116 CHALLENGES IN IMPLEMENTATION 1. Data Acquisition and Quality A primary challenge lies in the acquisition of reliable and consistent data from both a technical and user-experience perspective. For the physiological pipeline, relying on consumer-grade wearables (like a boAt smartwatch) or CSV uploads introduces significant variability. This data is often "noisy," subject to artifacts from movement, poor sensor contact, or differing sample rates. The Pandas processing scripts must be robust enough to handle missing data, outliers, and varying CSV formats. On the psychological side, the Web Speech API's accuracy is highly dependent on the user's microphone quality, background noise, and accent, which can lead to critical transcription errors (e.g., "I feel fine" being transcribed as "I feel pain"), thereby corrupting the sentiment analysis input. 2. Limitations of AI and NLP Models The "intelligence" of the system is constrained by the inherent limitations of its AI models. The VADER sentiment model, while fast, is lexicon-based and struggles with complex linguistic phenomena. It can be easily confused by sarcasm, context-dependent statements ("That was a sick presentation"), or nuanced emotional expressions, leading to a simplistic or incorrect emotional score. Similarly, the Goslate library, which relies on Google Translate, can fail to capture the subtle, empathetic nuances required for mental health communication. A literal translation of an empathetic English phrase might sound sterile, robotic, or even grammatically incorrect in the target language, potentially harming user rapport. 3. The Core Challenge: Multi-modal Data Fusion The most significant algorithmic challenge is the fusion of physiological and psychological data. This is an "apples and oranges" problem: how to meaningfully combine a VADER compound score (a float from -1.0 to +1.0) with an HRV metric (e.g., RMSSD in milliseconds). This process requires two critical, and largely subjective, steps. First, normalization, where both values must be scaled to a common range (e.g., 0 to 1). Second, weighting, where the system must decide the relative importance of each stream. For instance, should a highly negative text entry (high stress) override a calm physiological reading (low stress), or vice-versa? This weighting logic is difficult to define and even harder to validate without a large, labeled dataset. 4. Asynchronous Data Synchronization Closely related to data fusion is the challenge of temporal synchronization. The user's physiological state (from a CSV file) might represent their average stress from 9:00 AM to 12:00 PM, but their textbased journal entry might be written at 7:00 PM, reflecting a specific event. Combining this asynchronous data to produce a single "Wellbeing Score" for 7:00 PM is fundamentally flawed, as it correlates data points that are hours apart. A valid implementation would require a sophisticated algorithm to align these timestamps, perhaps by time-weighting the physiological data or only using data from a very recent window, which adds significant complexity to the fusion logic. 5. Architectural and Deployment Complexity Finally, the choice of Electron to package a Flask backend presents a significant technical hurdle. Bundling a full Python environment (including heavy libraries like Pandas) into a self-contained desktop application is notoriously difficult. This can lead to bloated application sizes, cross-platform dependency issues (e.g., a package working on Windows but failing on macOS), and slow startup times. Furthermore, International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 117 this architecture requires robust inter-process communication (IPC). The main Electron process must reliably start, monitor, and (if it crashes) restart the background Flask server, creating a layer of complexity not present in a traditional, monolithic web or native application. TESTING Testing is a critical phase in the development lifecycle of the Mental Wellbeing Monitoring System, ensuring its functionality, reliability, and adherence to specified requirements. Given the multi-modal nature and local-first architecture of the application, a comprehensive testing strategy encompassing unit, integration, system, and user acceptance testing (UAT) is essential. 1. Unit Testing Unit testing focuses on verifying the smallest testable parts of the application in isolation. This granular level of testing is performed primarily on the Python backend modules and key JavaScript functions. Backend (Python) Unit Tests: Individual functions within the Pandas data processing module (e.g., functions for calculating HRV features, handling missing CSV data), the VADER sentiment analysis module (e.g., Sentiment Intensity Analyzer().polarity_scores() on specific test strings), and Werkzeug Security password hashing/verification functions are tested independently. Database interactions through Flask-SQLAlchemy are also unit-tested, typically by mocking the database calls to ensure ORM logic is correct without actual database writes. Frontend (JavaScript) Unit Tests: Key JavaScript functions responsible for UI updates, API call formatting, and client-side data manipulation (e.g., Chart.js data preparation functions) are tested using frameworks like Jest or Mocha. This ensures that client-side logic performs as expected before integration. [Consider adding a small flowchart here: "Flowchart for Unit Testing Process" showing individual function -> test case -> pass/fail] 2. Integration Testing Integration testing verifies the interaction between different modules or services within the system. For this project, it is crucial due to the distinct pipelines and the communication between the frontend and backend. Frontend-Backend API Integration: Tests ensure that JavaScript fetch requests correctly send data to Flask API endpoints and that Flask responses are correctly received and parsed by the frontend. This includes testing user login/registration, text submission to the chatbot, and CSV file uploads. Pipeline Integration: The interaction between the Pandas processing and database storage, as well as the flow from user text input -> VADER -> Goslate -> Flask response, is thoroughly tested. This ensures that data flows correctly through each stage of both the physiological and psychological pipelines. Database Integration: Tests confirm that data is correctly written to and retrieved from the SQLite database via Flask-SQLAlchemy for all operations, including user profile updates, historical data storage, and retrieval for dashboard visualization. 3. System Testing System testing evaluates the complete, integrated application to ensure it meets all specified requirements. This type of testing mimics real-world usage scenarios. End-to-End Workflow Testing: Comprehensive test cases cover entire user journeys, such as: User registration -> login -> upload CSV -> chat with AI -> view dashboard with updated scores. International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 118 Testing the complete functionality of the "Wellbeing Score" fusion logic by providing specific physiological and psychological inputs and verifying the expected combined output. Performance Testing: While not a high-performance system, basic tests ensure responsiveness. For example, verifying that chatbot responses are generated within an acceptable timeframe (e.g., < 2 seconds) and that dashboard charts load quickly. Security Testing: Given the local-first privacy model, tests verify that no sensitive data is transmitted externally. This includes network monitoring during operations to confirm no unauthorized external API calls or data transfers occur. Password hashing integrity is also verified. Cross-Platform Compatibility Testing: The Electron package is installed and tested on target operating systems (Windows, macOS, Linux) to ensure consistent functionality, UI rendering, and performance across different environments. [Consider adding a pie chart here: "Pie Chart of System Test Case Distribution" showing percentages for: User Journey, Performance, Security, Cross-Platform] 4. User Acceptance Testing (UAT) UAT involves end-users (or representatives) testing the application in a realistic environment to validate that it meets their needs and expectations from a usability and functional perspective. Usability Testing: Users are asked to perform typical tasks (e.g., navigate the dashboard, interact with the chatbot, understand the wellbeing score). Feedback is collected on ease of use, clarity of information, and overall user experience. Functionality Verification: Users confirm that all features work as expected and that the recommendations provided by the AI chatbot are relevant and helpful. Satisfaction Surveys: Post-testing surveys are used to gather qualitative feedback on the application's perceived value, intuitiveness, and overall satisfaction. UAT is crucial for validating the effectiveness of the AI chatbot's responses and the interpretability of the wellbeing score. International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 125 The provided screenshots illustrate the main Dashboard of the MindWell application, acting as the central command center for user monitoring. The interface features a navigation sidebar for seamless access to modules like the Journal and AI Chat, while the main content area visualizes key health metrics. The "Upload Watch Data" section demonstrates the system's physiological pipeline in action, where a user has uploaded a CSV file containing wearable sensor data. Following this upload, the backend—powered by Pandas—has processed the raw input to generate immediate insights, which are displayed in the "Health Summary" and top-level cards. Specifically, the system reports an Average Heart Rate of 98.77 bpm and, by analyzing the heart rate variability within the file, has calculated and displayed a "Moderate" Stress Level. This setup effectively showcases the application's capability to ingest raw external data and instantly transform it into actionable, easy-tounderstand feedback for the user. This figure illustrates the Daily Journaling Module, a key component of the system's psychological data International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 126 pipeline designed to capture qualitative mental health insights. In this interface, the user inputs a free-text reflection on their day, which is then processed by the backend using the VADER sentiment analysis algorithm. Upon clicking "Analyze Emotions," the system evaluates the text's emotional valence— identifying positive keywords such as "felt right" and "genuinely rested"—and computes a quantitative Mood Score (displayed here as 9.98). The interface immediately feeds this analysis back to the user, categorizing the overall sentiment as "Positive" and providing encouraging, context-aware advice, thereby closing the loop between self-expression and AI-driven feedback . This figure depicts the Interactive AI Chatbot Interface, which serves as the system's primary intervention tool. The conversation history demonstrates the Natural Language Processing (NLP) pipeline in real-time: upon receiving the user's input "I am very depressed," the backend's VADER algorithm detects a highly negative sentiment valence. This triggers a logic-based response that prioritizes empathy ("It's okay to not be okay") and transitions into a practical recommendation, specifically the "4-7-8 Breathing" technique. Additionally, the interface highlights the system's accessibility features, including a microphone icon for voice interaction via the Web Speech API and a language selector (currently set to "English (India)") that leverages Goslate for multilingual support. International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 127 This figure displays the User Profile and Data History Interface, serving as the comprehensive log for longitudinal wellbeing tracking. It visually demonstrates the system's data persistence capabilities, retrieving stored records from the local SQLite database. The interface is segmented into two distinct data streams: the "Mood History" column, which lists chronological entries with their corresponding sentiment labels and quantitative scores (e.g., "Positive (Score: 9.98)") derived from the VADER analysis of past journals; and the "Wearable Data" column, which presents the physiological metrics processed from CSV uploads, including Step count, Heart Rate, and the calculated Stress Level. This view validates the system's multi-modal nature, proving that both psychological and physiological data are being successfully captured, analyzed, and archived for user review. This figure illustrates the Professional Intervention Module, specifically the "Find a Counselor" interface. While the MindWell application primarily leverages AI for immediate support and selfmonitoring, this module serves as a critical escalation pathway for users requiring clinical assistance. The interface features a clean, card-based directory of mental health professionals, presenting users with essential decision-making data including specializations (e.g., "Stress & Anxiety"), professional bios, patient ratings, and reviews. By providing a seamless route to connect with human experts, this feature ensures the system acts not just as a tracking tool, but as a comprehensive bridge between digital self-care and professional medical intervention. CONCLUSION This project has successfully designed and implemented a privacy-centric, multi-modal application for monitoring mental wellbeing. By bridging the gap between physiological signals and psychological states, the system addresses the limitations of traditional stress tracking methods, which often rely on isolated metrics. The fusion of Heart Rate Variability (HRV) data with VADER-driven sentiment analysis allows for a nuanced "Wellbeing Score" that correlates physical stress indicators with emotional context. Furthermore, the integration of the Web Speech API and Goslate ensures the system is accessible and interactive, lowering the barrier for users to engage in regular self-reflection. A significant contribution of this work is its architectural approach to data privacy. By packaging a Flask backend and SQLite database within a local Electron environment, the system proves that International Journal of Advanced Scientific and Technical Research ISSN 2249-9954 Available online on http://www.rspublication.com/ijst/index.html volume 15, No. 6, 2025 DOI: 10.5281/zenodo.17726758 Original Article ©2025 RS Publication, [email protected] 128 sophisticated health analytics can be performed entirely on the client side, eliminating the risks associated with cloud-based storage of sensitive medical data. This "local-first" design philosophy is increasingly critical in the modern digital health landscape. However, the study also acknowledges inherent limitations. The reliance on lexicon-based sentiment models and consumer-grade sensor simulations restricts the clinical precision of the current prototype. Future work will focus on integrating direct API feeds from wearable devices for real-time telemetry and upgrading the NLP engine to Large Language Models (LLMs) like BERT or GPT to capture deeper emotional nuances. Ultimately, this system serves as a scalable proof-of-concept, paving the way for future innovations in personalized, secure, and AI-assisted mental healthcare. REFERENCE [1] VADER Sentiment Analysis Paper: Hutto, C. J., & Gilbert, E. (2014). VADER: A Parsimonious Rule-based Model for Sentiment Analysis of social media Text. https://ojs.aaai.org/index.php/ICWSM/article/view/14550 [2] Multi-modal Data Fusion Paper: García-Ceja, E., et al. (2018). Mental Health Monitoring Using Multi-Modal Sensing: A Review. https://ieeexplore.ieee.org/document/8424843 [3] AI Chatbots in Mental Health Paper: Abd-Alrazaq, A. A., et al. (2019). An Overview of the Features of Chatbots in Mental Health: A Scoping Review. https://doi.org/10.1016/j.ijmedinf.2019.103978 [4] Physiological Stress & HRV Paper: Kim, H. G., et al. (2018). Stress and Heart Rate Variability: A Meta-Analysis and Review of the Literature. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5900369/ [5] Multi-Modal Data Fusion Citation: E. García-Ceja, S. Raurale, R. F. Brena, and C. Toma, "Mental Health Monitoring Using Multi-Modal Sensing: A Review," IEEE Access, vol. 6, pp. 45868-45885, 2018. https://ieeexplore.ieee.org/document/8424843 [6] Edge Computing & Privacy (IoT) Citation: M. Al-Khafajiy, L. Webster, T. Baker, and A. Waraich, "Smart Healthcare Monitoring using IoT and Edge Computing," 2018 11th International Conference on Developments in eSystems Engineering (DeSE), Cambridge, UK, 2018, pp. 283-288. https://ieeexplore.ieee.org/document/8647895