This project was a collective effort, where each member handled core features and critical components of TheraMind. Each team member devoted their full energy and commitment throughout the stages of project conceptualization, development, testing, and deployment.
- Abdullah Imran – Developed Authentication, TheraChat (AI Chatbot), Education Module, and Treatment Plan System across both frontend and backend. Trained and deployed all AI models in the backend, managed full project deployment, and contributed to dataset preparation for model development.
- Ambreen – Designed and implemented the Questionnaire & Diagnosis System, built the complex Appointment Booking & Management workflows, and developed the About Us and Contact Us pages. Also contributed to dataset preparation for AI model training.
- Hamda Qadeer – Created the Homepage, Meditation Module, and the comprehensive Admin Panel. Additionally, contributed significantly to dataset preparation for AI model training.
Final Year Project (2025) – University of Management and Technology, Lahore
TheraMind is a web-based mental health support platform that guides individuals from initiation to actionable treatment plans. It provides personalized mental health support across five major conditions — Stress, Trauma, Depression, Anxiety, and OCD — each with three subtypes.
Unlike traditional systems, TheraMind combines questionnaire-based diagnosis, appointment booking, treatment plan design, AI-driven chat support, educational content, and meditation exercises into one coherent ecosystem.
- Signup & complete a guided questionnaire
- Receive a personalized diagnosis & recommendations
- Book & attend doctor appointments via Google Meet
- Access a structured treatment plan designed by your doctor
- Explore educational content (articles, patient stories)
- Use meditation timers & guided videos
- Interact with TheraChat (AI-powered empathetic chatbot)
- Register & wait for admin approval
- Manage available appointment slots
- Conduct video appointments with patients
- Design & update treatment plans (goal-action structure with versioning & priority levels)
- Write & publish articles
- Use TheraChat for support and research
- Approve or block doctors
- Manage patient/doctor accounts
- Oversee system analytics (Firebase Analytics)
- Communicate with developers for escalations
-
Meditation → Breathing timers + curated YouTube video library
-
Education → Tag-based article & patient story browsing
-
Write → Patients submit stories, doctors publish articles (AI filters for relevance using CNN+SBERT)
-
Treatment Plans → Immutable, weekly versioned, with weighted actions (priority levels 1–3)
-
Questionnaire → Multi-stage diagnosis with suicide-risk detection & tailored condition-specific questions
-
Appointments → Google Meet integration, rescheduling, cancellation, and email reminders
-
TheraChat → Gemini-powered chatbot with custom fine-tuning for:
- Empathetic support
- Knowledge-based answers
- Refusal on irrelevant queries
flowchart LR
A[Patient/Doctor/Admin] -->|React + Tailwind| B[Frontend - Vercel]
B -->|REST API| C[Django Backend - EC2]
C --> D[Firestore DB]
C --> E[AI Models - TensorFlow/Keras/SBERT]
C --> F[Gemini API - Vertex AI]
C --> G[Google Services - OAuth, Calendar, Meet]
C --> H[EmailJS]
- React JS - Frontend framework
- Tailwind CSS - Styling and design
- Django - Backend framework
- Google Firestore - NoSQL database
- Google Firebase - Authentication
- AI/NLP → TensorFlow, Keras, NLTK, Hugging Face, SBERT + CNN
- Chatbot → Gemini API + Vertex AI endpoint
- Authentication → Google Identity Services (OAuth 2.0)
- Appointments → Google Calendar + Meet APIs
- Notifications → EmailJS
- Analytics → Firebase Analytics
| Component | Service Used | Notes |
|---|---|---|
| Frontend | Vercel | Hosts React + Tailwind app |
| Backend | Amazon EC2 | Django + AI Models deployed |
| Domain | Hostinger + NGINX | theramind.site → Vercel, api.theramind.site → EC2 |
| Models | EC2 (TensorFlow, SBERT + 1D-CNN) | Served via backend |
| Collaboration Tools | GitHub, ClickUp, Google Colab | VCS + Agile + Model Training |
We experimented with multiple architectures for content filtering & chatbot training:
- Initial Attempts: Logistic Regression, LSTM, GRU, BiLSTM → Encountered overfitting issues
- Final Solution: SBERT + 1D CNN (with negative mining + noise injection) → Achieved stable & accurate classification
| Class | Precision | Recall | F1-Score | Support |
|---|---|---|---|---|
| Non-MH | 0.95 | 1.00 | 0.97 | 109 |
| MH | 1.00 | 0.94 | 0.97 | 109 |
| Accuracy | 0.97 | 218 | ||
| Macro Avg | 0.97 | 0.97 | 0.97 | 218 |
| Weighted Avg | 0.97 | 0.97 | 0.97 | 218 |
- V1 → Over-restricted, failed to answer knowledge questions
- V2 → Retrained with expanded dataset → Achieved balanced empathy + knowledge (85.9% accuracy)
- Dependency Conflicts → TensorFlow & NLP libraries caused major conflicts, resolved via environment isolation
- Cloud Deployment → Models ran fine locally but failed in Docker/EC2 initially → resolved with optimized images
- Google OAuth Approval → Successfully navigated security checks for sensitive scopes
- AI Fine-tuning → Achieved balance between empathy & knowledge after multiple dataset iterations
git clone https://github.com/poetabdullah/theramind.git
cd theramindnpm install
npm startcd backend
venv\scripts\activate
pip install -r requirements.txt
python manage.py runserver- Firebase credentials
- Gemini API keys
- Google OAuth credentials
- EmailJS configuration
This project was developed as an academic final year project. Usage permissions can be discussed upon request.
- Live Demo: theramind.site
- Project Repository: GitHub
Built with ❤️ for mental health awareness and support