PROJECT PROPOSAL

for

Mind Mate AI: Mental Health Support Chatbot

Department of Artificial Intelligence and Data Science

Vishwakarma Institute of Information Technology,
Pune 411048

Submitted By

Harshita Kukreja - 22310999

Saloni Shahane – 22311814
Aditya Deshmukh - 22311670

1
 TABLE OF CONTENTS

SR.NO. CONTENT PAGE NO.

1. Problem Statement 3

2. Objectives 3

3. Data Collection 3

4. AI/ML Algorithms & Methodology 4

5. System Architecture 5

6. Expected Outcomes 6

7. Ethical Considerations 7

8. Timelines & Milestones 7

9. Approximate Budget 8

10. Literary Survey 8

11. Conclusion 10

2
 1. PROBLEM STATEMENT:
Many individuals struggle with mental health support due to stigma, lack of accessibility, or the
need for different interaction styles. MindMate AI addresses this by offering a dual-mode chatbot
—Professional Mode for structured guidance and solutions, and Friend Mode for casual,
judgment-free conversations. Users start with a smart questionnaire to identify concerns, choose
their preferred mode, and interact via voice or text. A dashboard provides a daily quote, chat
history, and an optional To-Do List for solution tracking.

2. OBJECTIVES:
Personalized Mental Health Support – Provide users with a tailored chatbot experience by
offering both Professional Mode (structured guidance) and Friend Mode (casual, judgment-free
conversations).

Smart Issue Identification – Utilize a questionnaire-based evaluation to understand user

concerns and suggest relevant sub-problems for better-targeted support.

Dual Interaction Modes – Allow users to switch between Professional Mode, where structured
guidance and solutions are provided, and Friend Mode, where users can vent freely or engage in
casual conversations.

Action-Oriented Approach – Enable users to convert solutions into a To-Do List (if opted),
which is displayed on their dashboard for easy tracking and self-improvement.

Multi-Modal Communication – Support both voice and text inputs to ensure accessibility and
ease of use for diverse users.

Continuous Engagement & Motivation – Provide a daily motivational quote and chat history
tracking on the dashboard, encouraging consistent mental wellness practices.

3. DATA COLLECTION:
For MindMate AI, a manually curated dataset has been developed to ensure accurate and
contextually appropriate responses in both Professional Mode and Friend Mode. The dataset is
structured into three columns:

• User Input – A diverse set of user statements covering a spectrum of mental health
concerns, ranging from basic everyday struggles to moderate and severe emotional
distress.

3
 • Friendly Response – A conversational, supportive, or humorous reply designed for
Friend Mode, where the chatbot interacts like a virtual companion.
• Professional Response – A structured, solution-oriented response tailored for
Professional Mode, providing guidance, coping strategies, or reflective questions to help
users navigate their challenges.

The dataset has been carefully designed to ensure inclusivity across various emotional states and
mental well-being concerns. By incorporating user inputs from mild stressors (e.g., feeling tired,
unmotivated) to more severe issues (e.g., anxiety, depression, burnout), the chatbot can adapt its
response style to suit the user’s needs effectively.

Moving forward, the dataset will be split strategically: - The Professional Responses will be
used to train the Mistral 7B model to generate structured, helpful replies. - These generated
professional outputs will then be passed through the Gemini API, which will convert them into
friendly, conversational versions suitable for Friend Mode.

This modular approach ensures both consistency and flexibility while allowing us to maintain
high response quality in both interaction styles. Over time, the dataset will continue to evolve
based on real user interactions and feedback, improving personalization and relevance in every
conversation.

This is the link to our dataset.

https://docs.google.com/spreadsheets/d/
1snnPtNhKkyHUefukBJQoR70CY2xZzEBPeEZo1HCBNQA/edit?usp=sharing

4. AI/ML ALGORITHM AND METHODOLOGY:

The platform leverages AI and ML to provide personalized mental health support across
different conversation modes. The system uses a combination of open-weight large language
models, tone adaptation APIs, sentiment analysis models, and productivity/motivation APIs to
create a holistic experience.

Mistral 7B Model

The core AI model used is Mistral 7B, an open-weight transformer-based large language model.
It is responsible for generating professional, structured, and informative responses to user
queries. These responses are used especially in the “Doctor” conversation mode, where a more
serious and expert tone is expected.

Gemini API for Tone Adaptation

After the Mistral 7B model generates a professional response, the Gemini API is used to rewrite
or adjust the tone of the response according to the selected conversation mode. For example:

• If the user selects the Friend mode, the response is rewritten in a friendly, comforting,
and casual tone.
4
 • If the Counselor mode is selected, the response is rewritten in a balanced, empathetic, and
self-help style.

Emotion Detection and Sentiment Analysis

A lightweight sentiment analysis model (such as a fine-tuned transformer model) is used to

detect emotional cues from user input. It identifies emotions like sadness, anger, anxiety, or
distress. Based on this detection:

• The chatbot adjusts its response tone.

• It logs the emotion in the user’s mood tracking history.
• If the sentiment indicates a crisis or distress, the chatbot activates the crisis mode,
suggesting professional help or providing helpline information.

To-do-list API Integration

The Todoist API is used to generate and manage personalized to-do lists for users. Based on user
interactions or suggestions by the bot, tasks can be created and synced to their Todoist account.
This promotes daily structure and productivity for mental wellness.

ZenQuotes API for Daily Motivation

The platform uses the ZenQuotes API to fetch one motivational quote each day, which is
displayed on the user’s dashboard. This small but consistent interaction supports daily positive
reinforcement and emotional well-being.

Together, these components form the AI/ML backbone of the system, enabling smart, safe, and
context-aware conversations.

5. SYSTEM ARCHITECTURE:
The system is built with a modular, scalable architecture that integrates AI models, third-party
APIs, and front-end and back-end technologies to deliver real-time mental health support.

Frontend (User Interface)

The frontend is developed using HTML, CSS, and JavaScript (or optionally a framework like
React). It includes:

• A chat interface supporting both text and voice input.

• Mood check-in and tracking interface.
• Display of daily motivational quotes.
• A personalized dashboard showing to-do tasks and emotion logs.

Backend (Application Server)

5
The backend is responsible for all server-side logic. It is built using a framework like Flask,
Django, or Node.js. Its core functions include:

• Handling user input and requests.

• Interacting with the Mistral 7B model to generate base responses.
• Calling the Gemini API to transform responses based on conversation mode.
• Analyzing input with the sentiment model for emotional state detection.
• Communicating with the Todoist API to manage tasks.
• Fetching motivational quotes from the ZenQuotes API.
• Storing session and mood tracking data securely.

AI Components

• Mistral 7B: Handles all core natural language understanding and generation.
• Gemini API: Adjusts the tone of Mistral’s responses to suit the Friend or Counselor
modes.
• Sentiment Analysis Model: Analyzes user text to determine emotional states and triggers
mood tracking or crisis alerts.

External APIs

• To-do-list API: Integrates with user accounts to create and manage self-care and
productivity tasks.
• ZenQuotes API: Provides daily motivational quotes displayed on the user’s dashboard.

Database: MongoDB

Used to store mood logs, chat history, to-do lists, and user preferences (anonymously if
required).

6. EXPECTED OUTCOMES:
The AI-powered mental health chatbot platform aims to deliver tangible, measurable, and user-
centered outcomes. By integrating intelligent conversational models, emotional understanding,
and wellness tools, the following results are anticipated:

Personalized Mental Health Support

Users will receive responses tailored to their emotional state and selected conversation mode
(Friend, Counselor, or Doctor), creating a more human-like and comforting interaction
experience.

Improved User Emotional Awareness

Through daily mood check-ins and sentiment detection, users will gain a better understanding of
their emotional patterns over time. This can support self-reflection and early recognition of
potential mental health concerns.

6
Productivity and Routine Enhancement

The integration of to-do list generation via Todoist will encourage users to maintain daily
structure, which is known to positively impact mental wellness. Personalized self-care or
wellness tasks will also enhance engagement.

Safe Space for Expression

The platform enables anonymous and secure conversations, encouraging users to open up
without fear of judgment. This outcome is critical in reducing barriers to seeking emotional
support.

Daily Positive Reinforcement

Displaying motivational quotes each day helps users start their day on a positive note, fostering
consistent emotional encouragement and optimism.

Proactive Crisis Support

The system’s ability to detect distress and activate crisis mode ensures that users showing signs
of emotional breakdown are not left unsupported. Providing helpline suggestions can help guide
them toward professional assistance when needed.

Dynamic and Adaptive Conversations

By using a combination of Mistral 7B and Gemini, the chatbot can shift its tone and depth
according to the user’s need—whether casual, self-help, or professional—offering a dynamic and
flexible support system.

Data-Driven Insights (Optional for Advanced Users)

Users who opt-in can benefit from emotion tracking over time, helping them monitor progress or
identify recurring triggers or patterns.

Engagement and Retention through Simplicity

A clean, conversational interface with minimal steps for accessing help ensures users return
regularly, enabling habit formation around self-care and emotional check-ins.

7. ETHICAL CONSIDERATIONS:
Given the sensitive nature of mental health, it is critical to address ethical responsibilities in the
design, development, and deployment of this AI-powered mental health platform. The following
considerations guide the system’s ethical framework:

User Privacy and Anonymity

The platform ensures users can access mental health support without revealing personal
information. Nicknames or anonymous logins are used to maintain privacy. No personally
identifiable information (PII) is collected unless explicitly permitted by the user.

7
Data Security and Confidentiality

All user inputs, mood logs, and conversation history (if stored) are handled securely using
encrypted communication and secure database storage. Access to sensitive data is restricted, and
data is never shared with third parties.

Crisis Management and Escalation

The system includes basic mechanisms for identifying signs of emotional distress. If detected,
the chatbot offers resources such as helpline numbers or encourages the user to seek help from a
licensed therapist. It avoids making critical decisions on behalf of the user.

Bias and Fairness in AI Responses

AI models like Mistral 7B and Gemini are trained on large datasets and may reflect societal
biases. The system includes prompt filtering and tone moderation to minimize biased,
insensitive, or harmful content. Continuous evaluation and feedback loops are established to
improve fairness over time.

Emotional Safety

The chatbot avoids triggering language, offensive content, or emotionally harmful suggestions.
Emotional safety is prioritized through careful tone management and sentiment-aware responses.

8. TIMELINES & MILESTONES:

The development of the AI-powered mental health chatbot platform was planned and executed
over a period of three weeks, divided into structured milestones to ensure consistent progress and
feature integration.

Week 1: Planning and Initial Setup

During the first week, the team focused on understanding user needs and finalizing the core
feature list. The UI wireframes were created, and the development environment was set up. APIs
for Gemini, ZenQuotes, and Todoist were tested for integration feasibility. A basic frontend
interface for chatbot interaction (both text and voice-based) was built. Ethical guidelines and data
privacy principles were also documented.

Week 2: AI Integration and Emotion Detection

The second week focused on backend development and AI integration. The Mistral 7B model
was used to generate professional, counselor-style responses, while the Gemini API was
integrated to convert these into friendlier, casual responses. Emotion detection based on user

8
prompts was implemented, along with the mood tracking and self-assessment modules. A simple
anonymous login system using nicknames was added for user privacy.

Week 3: Final Features and Testing

In the final week, all remaining features were completed and tested. The crisis mode was
activated to detect distress signals and respond with helpline suggestions. The dashboard was
updated to show daily motivational quotes from ZenQuotes and personalized task suggestions
using the TodoList API. The chatbot’s three conversation modes (Friend, Counselor, and Doctor)
were finalized. Extensive testing and debugging were done to ensure emotional tone consistency,
smooth user experience, and secure data handling. The final version was successfully deployed.

9. APPROXIMATE BUDGET
Computational Costs (Fine-tuning & Inference):
 Using Kaggle/Colab Pro → ₹1,000 - ₹5,000

Data Storage Cost:

 Storage (Google Drive or cloud storage) might cost ₹1,000 - ₹3,000/month

API & Hosting Costs:

₹Hugging Face Spaces (Free for small models, ₹3,000+ for higher usage)
₹AWS/GCP (₹5,000 - ₹15,000/month for a dedicated instance)
₹On-premise Hosting (₹0 if using your own PC, but electricity costs apply)

Miscellaneous Costs

 Domain Name (₹500 - ₹1,500/year)

 Website Hosting (₹3,000 - ₹10,000/year for a good server)
 UI/UX Design: ₹5,000 - ₹10,000
 Logo & Branding: ₹3,000 - ₹7,000

Total Estimated Costs:

 One-time Costs: ₹16,500 - ₹61,500

 Monthly Costs: ₹9,000 - ₹33,000

10. LITERARY SURVEY:

 Denecke, K., Abd-Alrazaq, A., Househ, M. (2021). Artificial Intelligence for
Chatbots in Mental Health: Opportunities and Challenges. In: Househ, M., Borycki,
E., Kushniruk, A. (eds) Multiple Perspectives on Artificial Intelligence in Healthcare.

9
 Lecture Notes in Bioengineering. Springer, Cham. https://doi.org/10.1007/978-3-030-
67303-1_10

From this paper, I learned that AI-powered chatbots hold significant potential in
supporting mental health care by offering accessible, scalable, and personalized
assistance. However, their effectiveness is limited by challenges such as data privacy,
ethical concerns, and the need for robust validation.

 Yue Qi. (2025) Pilot Quasi-Experimental Research on the Effectiveness of the

Woebot AI Chatbot for Reducing Mild Depression Symptoms among Athletes.
International Journal of Human–Computer Interaction 41:1, pages 452-459.

From this paper, I learned that the Woebot AI chatbot showed promising results in
reducing mild depression symptoms among athletes, highlighting its potential as a
supportive mental health tool. The study provides early evidence of its
effectiveness through a quasi-experimental approach.

 Fulmer R, et al. “Using Psychological Artificial Intelligence (Tess) to Relieve

Symptoms of Depression and Anxiety: Randomized Controlled Trial.” JMIR
Mental Health. 5.4 (2018): e64–e64. Web.

Psychological AI chatbot Tess significantly helped reduce symptoms of

depression and anxiety in users. The randomized controlled trial demonstrated its
effectiveness as a scalable mental health intervention.

 Wenwei Luo, Xiaoyu Wu, Ilene R. Berson, Michael J. Berson, Huihua He,
Minqi Gao, Exploring Technology8 Integration in Health and Safety Routines in
a Shanghai Kindergarten, Healthcare, 10.3390/healthcare13030218, 13, 3,
(218), (2025).

Primary focus is on health and safety routines in kindergartens, it also emphasizes

how technology integration can indirectly support young children's mental health
by creating a structured, secure, and responsive environment that reduces
stress and anxiety.

11. CONCLUSION:

10
This AI-powered mental health chatbot serves as a compassionate and intelligent support system
designed to make emotional wellness more accessible and personalized. Leveraging advanced
models such as Mistral 7B and the Gemini API, the platform offers real-time conversations that
adapt to the user’s emotional tone and needs. With features like multi-mode interaction
(including professional and friendly tones), emotion detection, anonymous chatting, and mood
tracking over time, the chatbot creates a safe space where users - especially GenZ can express
themselves freely without fear of judgment.

By incorporating context-aware responses and natural language understanding, the chatbot not
only listens but also responds in ways that feel human and empathetic. Whether someone is
navigating stress, loneliness, or just needs a comforting conversation, the system tailors its
replies to offer encouragement, clarity, and gentle guidance. Though it is not intended to replace
certified mental health professionals, it acts as a readily available companion for early emotional
support, daily check-ins, and self-reflection. Prioritizing privacy, ethical AI use, and mental well-
being, this chatbot bridges the gap between technology and mental health, making self-care more
intuitive, responsive, and stigma-free.

11