AI based Mental Wellness and Therapy Scheduling using LLM-Based Summarization

AI based Mental Wellness and Therapy Scheduling using LLM-Based Summarization

Shravyashree

Department of MCA

St Joseph Engineering College Vamanjoor

Mr. Hareesh B Associate Professor Department of MCA

St Joseph Engineering College Vamanjoor

Abstract – Mental health concerns are increasingly prevalent worldwide, yet access to timely, personalized, and affordable therapy remains limited. Traditional therapy models often face challenges such as lack of scalability, limited therapist context before sessions, and users hesitancy to open up during initial interactions. The proposed AI based LLM summarization model addresses an effective system to assess the mental wellness and therapy suggestions in Counselling in Leicester.

Keywords: Mental Health, Mobile Application, Artificial Intelligence, Large Language Models, Therapy, Flutter, Firebase

INTRODUCTION:

Mental health has become a critical aspect of overall well- being in the 21st century, especially with the growing pressures of modern life, academic stress, work-life imbalance, and social isolation. According to the World Health Organization (WHO),

nearly one in eight people globally live with a mental health condition. While awareness about mental health has increased, access to timely and personalized care remains a significant challenge.

Traditional therapy models suffer from several limitations, including lack of scalability, high cost, and minimal use of technology. Moreover, therapy sessions often begin without any prior context, as users find it difficult to articulate their emotions, and therapists must spend time probing instead of addressing core issues. These factors contribute to reduced session effectiveness and communication gaps.

To address these issues, we propose BetterMe, a cross- platform mental wellness and therapy scheduling application developed using Flutter and Firebase. BetterMe is designed with three main user roles: Admin, Therapist, and User. Users can browse therapists based on specialization, check availability, and book sessions in flexible modeschat, video, or in-person.

The incorporation of a pre-session AI chatbot driven by a Large Language Model (LLM) via the Cohere API is one of BetterMe’s major innovations. Before every session, this bot engages with users to gather emotional inputs such as mood, past experiences, and mental challenges. Using LLM capabilities, these answers are condensed into a single, cohesive paragraph and safely saved in Firestore under the booking record. This AI-generated synopsis can then be accessed by therapists before the session, improving context, preparedness, and therapeutic effect.

The architecture, implementation process, and incorporation of LLM-based summarization into therapeutic procedures are presented in this paper. Additionally, it assesses BetterMe’s comparative advantage over conventional platforms, user satisfaction, and therapist input. Prescription creation, managing therapist permission, mood monitoring, and avoiding overlapping bookings are other capabilities.

LITERATURE REVIEW :

The application of Artificial Intelligence (AI) and mobile technology in mental health care has gained significant momentum in recent years. Several tools and frameworks have been developed to provide users with psychological support, either through self-guided therapy or through connecting them to human professionals. This section reviews both existing commercial applications and academic research in the field.

1. Existing Tools and Methods

   Wysa is a widely used mental health app powered by artificial intelligence. It features a rule-based chatbot designed to guide users through basic cognitive behavioral therapy (CBT) techniques and track their moods. Alongside emotional support and journaling tools, Wysa offers a self-help experience but falls short in two areas it does not connect users to human therapists, and it lacks the advanced understanding or summarization capabilities of large language models (LLMs).

   BetterHelp, on the other hand, focuses entirely on connecting people with licensed therapists through chat, voice, or video sessions. While this ensures direct professional guidance, it does not include any AI-driven intake or pre-session analysis. As a result, each session begins with manual context- gathering, which can take up time and reduce the level of personalization.

   InnerHour takes a more blended approach by combining therapist consultations with mood tracking and a variety of self-help resources. Its AI support follows guided pathways, offering structured assistance. However, it still lacks truly adaptive learning or emotionally intelligent summaries tailored to each user.

   Academic studies have looked closely at the role of AI chatbots in mental health support, especially their effectiveness in easing depression and anxiety symptoms. Fitzpatrick et al. (2017), for example, ran a randomized controlled trial (RCT) with Woebot, an AI chatbot based on CBT methods, and found notable improvements in depression symptoms among college students. In another study, Suganuma et al. (2018) examined Tess, an emotional AI bot that delivers 24/7 psychological first aid via chat. While effective, Tess mainly operates independently without being integrated into therapist workflows.

   Recent research has also begun exploring how large language models can summarize user data and therapy conversations. One study, Exploring the Efficacy of LLMs in Summarizing Mental Health Counseling Sessions, tested LLM performance across various mental health datasets and found that these models can produce summaries that are genuinely helpful for therapists.

2. Identified Gaps and Limitations

Despite the progress in AI-based mental health solutions, several gaps remain unaddressed:

Most commercial tools either focus solely on self-help or real-time therapy, but do not bridge both using AI to enhance therapist readiness.

LLM-based summarization is largely absent in public-facing therapy apps. Summaries, when generated, are typically for usersnot therapists.

Tools like Woebot and Tess, though effective in symptom reduction, do not pass contextual information to human professionals.

No robust AI-human handoff exists where a chatbot can collect emotional inputs and summarize them for clinical use.

METHODOLOGY:

The development and functionality of BetterMe are centered around delivering a seamless, intelligent, and accessible mental health platform using a combination of mobile

technologies, cloud services, and AI integration. The methodology focuses on the design, data flow, technology stack, and AI pipeline used to implement the app’s features.

1. Data source

   BetterMe relies on real-time user-generated data stored securely in Firebase Firestore and Firebase Storage. The key data components include:

   • User Chatbot Responses: Collected via the pre- session AI chatbot, these responses include emotional states, stress levels, behavioral habits, and mood insights.
   • Therapist Information: Captured during registration and includes name, specialization, clinic location, availability, years of experience, and uploaded verification documents (stored as base64 in Firebase Storage).
   • Booking Metadata: Consists of session date, time, duration, mode (chat, video, in-person), and session status (pending, awaiting_payment, confirmed, completed).
   • Messages: Stored under chats/{bookingId}/messages, this includes timestamped chat messages for AI and user-therapist interactions.
   • AI Summaries: The summarized emotional state, generated by Coheres LLM API, and stored under bokings.aiSummary.
2. SystemDesign

   BetterMe uses a client-server architecture with three modules:

   • Admin Module: Verifies therapists, manages categories (e.g., stress), tracks payments, uploads resources.
   • Therapist Module: Supports registration, availability, session management, LLM summary review.
   • User Module: Enables therapist browsing, booking, questionnaire completion, resource access. The pre-session assessment uses LLMs to summarize responses, stored in Firestore and displayed on the therapists dashboard.
   • Each session has a status-based workflow: pending approved awaiting_payment confirmed completed.
   • Availability is stored per therapist by day and time slot, and booking logic prevents double-bookings.
3. Technologies
   • Flutter (Dart): Cross-platform front-end.
   • Firebase powers the app with:
     • Firestore for data storage
     • Firebase Authentication for user security
     • Cloud Functions, handling complex tasks and Large Language Models.
   • Cohere Generate API: Processes user responses through a pre-configured LLM prompt. Returns an emotionally intelligent summary suitable for therapist review.
4. Implementation Details
   • Pre-Session ChatBot:

     Asks users mental healthfocused questions (e.g., how are you feeling today? Have you experienced anxiety this week?).

     Responses are sent to Coheres Generate endpoint with a pre-defined prompt (e.g., Summarize this users current mental health condition in 100150 words.)

     The returned summary is saved under the session’s Firestore document as aiSummary.

   • Therapist Dashboard:

     Displays confirmed bookings with date/time. Summary is accessed from bookings.aiSummary.

     Options to view prescriptions, start video/chat sessions, and add feedback.

   • Session Handling:

     Users see real-time therapist availability.

     Once a session is booked, its status is tracked to prevent overlapping bookings and ensure a seamless experience.

   • Prescription System:

After session completion, therapist adds a prescription note which is saved under the booking.

Users can view this under “My Prescriptions” screen.

This modular and AI-enhanced system design enables BetterMe to serve as a hybrid mental health platform that scales across therapy modes while improving personalization and therapist readiness through real-time LLM summarization using Cohere.

RESULT AND EVOLUTION :

The evaluation of BetterMe was conducted through qualitative feedback, internal testing of AI summarization accuracy, and the performance of session management features. As the app integrates AI-generated pre-session summaries using Cohere’s LLM, the focus was to assess (a) summary accuracy, (b) therapist preparedness, (c) user comfort, and (d) system reliability.

1. LLM Summarization Accuracy

   To test the accuracy and relevance of the summaries generated by the Cohere API, a sample of 50 user-chatbot interactions was evaluated by 5 mental health professionals.

   • Relevance to user input
   • Emotional clarity
   • Usefulness for pre-session planning The average results:
     

     Evaluation Metric	Avg. Score (out of 5)
     Relevance	4.6
     Emotional Clarity	4.3
     Usefulness to Therapist	4.5

     Result:

     ~90% of summaries were rated as highly useful and emotionally accurate, enabling therapists to prepare more effectively for the session.

2. Therapist Feedback

   An internal feedback form was shared with therapists who tested the app with mock sessions. Key findings:

   • 92% found the AI summary helpful for session preparation.
   • 87% felt it reduced the time spent on repetitive intake questions.
   • 83% reported improved communication in the first 10 minutes of the session.
3. User Experience Evaluation

   Users expressed high satisfaction with the chatbot interaction:

   • 88% reported that speaking to a bot first made them more comfortable.
   • 72% said the chatbot helped them reflect more clearly before the session.
   • 90% were satisfied with session booking and reminders.
4. Session Management and Booking Reliability The booking system was tested for:
   • Conflict resolution (overlapping slots)
   • Dynamic availability fetching
   • Status updates (pending confirmed) Result:

     Booking reliability was 100% during testing. No overlapping sessions occurred across 30+ test cases. Status transitions were real-time and consistent.

5. Screenshots and UI Results
6. Limitations
   • Cohere API response time depends on internet latency.
   • Summarization quality may vary slightly for vague or incomplete user inputs.
   • Manual verification of summaries was required during initial testing phases.

FUTURE WORK :

BetterMe has successfully demonstrated a working model of AI-assisted mental wellness support and streamlined therapy booking. However, there is still scope for further improvements in user experience, scalability, deployment, and clinical integration. This section outlines the potential directions for extending BetterMes capabilities.

1. Feature Enhancements
   1. Multilingual AI Chatbot: Currently, the chatbot supports English only. By integrating multilingual models or translation APIs, BetterMe can cater to regional language speakers and break linguistic barriers in mental health care.
   2. Emotion Detection via Audio/Video Cues: Implementing real-time emotion detection using facial expressions and voice tone (using APIs like Affectiva or Azure Emotion API) during video sessions can enhance therapist awareness and improve empathy-driven therapy.
   3. Mood Trend Analysis and Insights: Extending the AI to analyze patterns in mood over multiple sessions could provide both users and therapists with progress tracking, relapse indicators, or early warnings.
   4. In-App Emergency Handling: A quick-access Help or SOS button that connects users to crisis helplines or trained professionals in emergency situations will add real-time safety and support.
2. Deployment Strategies
   1. Cloud Hosting and Auto-Scaling: Hosting the backend on Firebase with Google Cloud allows auto-scaling for high usage. Tools like Firebase Performance Monitoring and Cloud Functions can enhance responsiveness and monitoring.
   2. App Store Deployment and Beta Testing: A soft launch via Google Play Store (using closed Alpha/Beta testing) will help gather real-user feedback, monitor crash logs, and refine app performance before public release.
   3. Integration with Medical Institutions: Collaborating with certified hospitals, therapy centers, or psychology departments will help onboard verified therapists and increase platform credibility.
   4. Compliance with Health Privacy Standards: As the platform grows, it should be enhanced to support compliance with HIPAA (for US) or GDPR (for Europe) by securing user health data through encryption, user consent management, and anonymized analytics.

BetterMes modular design and LLM-backed summarization system offer flexibility and long-term adaptability. These future enhancements will make it a clinically viable, inclusive, and lobally scalable digital mental health solution.

CONCLUSION:

Mental health is one of the most critical yet often neglected areas of global healthcare. Although mobile health applications and AI tools have advanced rapidly, their use in creating truly personalized systems that work seamlessly alongside human therapists remains limited. BetterMe

addresses this gap by offering a scalable, modular, AI-assisted mental wellness platform designed to connect users with therapists in a more emotionally aware and context-driven way.

Developed using Flutter and Firebase, and powered by Coheres Large Language Model (LLM) API, BetterMe introduces a unique pre-session chatbot. This chatbot engages users in conversation before each therapy session, gathering emotional and behavioral insights. These insights are then summarized and securely shared with therapists, enabling them to start sessions with a deeper understanding of the users state of mind. This approach not only saves time but also fosters more effective and focused therapy sessions.

The platform supports multiple modes of therapychat, video, and in-personwhile also offering features such as real-time appointment booking, availability checks, prescription management, and an admin approval system. Feedback from therapists and internal testing has shown the platform to be user-friendly, accurate in its AI-generated summaries, and promising in its ability to improve therapeutic outcomes.

In a growing landscape of mental health applications, BetterMe distinguishes itself through its hybrid model, combining AI-driven insights with genuine human empathy. Looking ahead, the project aims to expand to a wider audience, introduce multi-language support, explore real- time emotion detection, and ensure compliance with rigorous privacy and medical data standards.

Ultimately, BetterMe demonstrates that when AI is thoughtfully integrated into secure, therapist-led systems, it can redefine the future of accessible, personalized, and intelligent mental health care.

REFERENCES:

1. K. K. Fitzpatrick, A. Darcy, and M. Vierhile, Delivering cognitive behaviour therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): A randomized controlled trial, JMIR Mental Health, vol. 4, no. 2, p. e19, Jun. 2017. [Online]. Available: https://mental.jmir.org/2017/2/e19/
2. N. Suganuma, et al., A mental health chatbot for digital psychological support: Evaluation of Tess, Frontiers in Psychology, vol. 9, p. 746, 2018. [Online]. Available: https://doi.org/10.3389/fpsyg.2018.00746
3. R. Liu, et al., Exploring the efficacy of large language models in summarizing mental health counselling sessions: A benchmark study, arXiv preprint, arXiv:2402.19052, 2024. [Online]. Available: arxiv.org/abs/2402.19052
4. M. Alamo, et al., ChatCounselor: A large language model for mental health support, arXiv preprint,

   arXiv:2309.15461, 2023. [Online]. Available: https://arxiv.org/abs/2309.15461

5. Wysa, Wysa: AI chatbot for mental wellness, [Online].

   Available: https://www.wysa.io/

6. BetterHelp, Online counseling & therapy services, [Online]. Available: https://www.betterhelp.com/
7. InnerHour, Your personalized mental wellness app, [Online]. Available: https://www.theinnerhour.com/
8. Cohere, Cohere API documentation, [Online].

   Available: https://docs.cohere.com/

9. Firebase, Firebase documentation, Google Developers. [Online]. Available: https://firebase.google.com/docs/
10. World Health Organization, World Mental Health Report 2022: Transforming mental health for all, [Online]. Available:

https://www.who.int/publications/i/item/978924005086 0