SRM INSTITUTE OF SCIENCE AND

TECHNOLOGY
SCHOOL OF COMPUTING
DEPARTMENT OF COMPUTING TECHNOLOGIES
18CSP109L- MAJOR PROJECT

Mental Health Support

Chatbot Using GenAI

Student 1 Reg. No: RA2111003011804

Guide name: Dr. Ponmagal R S Student 1 Name: Harsh Deep
Associate ProfessorDepartment of Computing
Technologies Student 2 Reg. No: RA2111003011797
Student 2 Name: Devesh Yadav
 Abstract
Introduces a generative AI-powered mental health support chatbot
designed to provide secure, empathetic, and accurate assistance.
Built using the ChatGPT API, leveraging prompt engineering,
LangChain refinement, and advanced sentiment analysis for
dynamic responses.
Integrates Pinecone Vector database and cosine similarity
algorithms for precise data retrieval and query resolution, powered
by OpenAI models.
Ensures security and accessibility through API key encryption and
moderation mechanisms for a positive user experience.
Connects users to nearby support centers and emergency
services, detecting self-harm tones and alerting authorities when
necessary.
Combines innovation, safety, and efficiency to address mental
health needs effectively.
 Introduction
Highlights the importance of mental health as a critical aspect
of overall well-being and addresses challenges in accessing
timely support.
Aims to bridge the gap by developing an AI-powered chatbot
that provides reliable mental health assistance.
Utilizes the ChatGPT API to deliver empathetic responses,
accurate information, and access to resources.
Features prompt engineering, sentiment analysis, and advanced
natural language processing for tailored interactions.
Focuses on security with a secure vault mechanism for API key
protection and moderation for a safe user environment.
Integrates APIs to connect users with nearby support centers
and emergency services, extending support beyond virtual
conversations.
 Existing System
1. Traditional Helpline Services
Provide direct communication with counselors or therapists.
Limited by factors like availability, long wait times, and user
hesitation due to stigma.
Lack immediate response during critical situations in some cases.
2. Digital Mental Health Solutions
Include apps and chatbots like Woebot and Wysa that offer AI-
driven interactions.
Focus on stress and anxiety management through basic
conversational support.
Often lack advanced features such as:
Personalized and dynamic responses using sentiment analysis.
Integration with real-world services like emergency support
centers.
 Problem Statement
Addresses the growing concern of mental health issues and the limited access to timely,

reliable, and stigma-free support.

Identifies challenges with traditional helpline services, including availability issues and delays.

Highlights limitations of existing digital mental health solutions, such as lack of personalization,

advanced sentiment analysis, and real-world emergency service integration.

Emphasizes concerns over data security and accuracy in query resolution, impacting the

effectiveness of current solutions.

Proposes a secure, feature-rich AI-powered chatbot offering empathetic responses, accurate

information, and integration with support centers and emergency services.

Aims to deliver a comprehensive mental health support solution addressing these critical

challenges.
 Objectives
Develop an AI-Powered Chatbot
- Leverage the ChatGPT API to create a responsive, empathetic chatbot for mental health support.

Enhance User Interaction with Sentiment Analysis

- Implement sentiment analysis to adapt the chatbot’s responses based on the user's emotional state.

Improve Response Accuracy

- Use prompt engineering and LangChain for faster and more accurate responses through refined
conversational models.

Integrate Secure Data Handling

- Implement encryption and security measures for API key protection and user data privacy.

Provide Emergency and Support Center Integration

- Incorporate APIs to connect users with nearby support centers and emergency services, especially for
self-harm detection.

Moderate Content and Ensure Safety

- Implement moderation features to filter out harmful or hateful content and create a safe environment
for users.

Offer Personalized Assistance

- Enable dynamic adjustments in the chatbot’s responses based on individual user needs and queries.
 Proposed System
AI-Powered Chatbot Framework - Uses the ChatGPT API for natural,
empathetic conversations and enhanced dialogue flow with GPT Turbo 3.5.
Sentiment Analysis and Personalization - Analyzes user sentiment to adjust
responses dynamically for a more personalized and compassionate interaction.
Prompt Engineering for Accuracy and Efficiency - Employs shot prompting
and LangChain to refine and speed up response generation.
Data Security and Encryption - Implements secure vault mechanisms to
protect API keys and ensure user data privacy.
Advanced Data Retrieval with Pinecone Vector Database - Utilizes
Pinecone’s vector database to perform cosine similarity searches for accurate
data retrieval.
Emergency Service and Support Center Integration - Connects users to local
support services and detects self-harm signals, alerting authorities when needed.
Moderation and Safety Features - Filters out harmful content to ensure a safe
and respectful environment for users.
Real-Time Assistance and Resources - Provides immediate responses with
tailored mental health resources and coping strategies.
Literature Review
 References
Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavioral therapy to young
adults with depression and anxiety via a smartphone application: A randomized controlled trial. Journal
of Medical Internet Research, 19(9), e307. https://doi.org/10.2196/jmir.7332
Poria, S., Hazarika, D., Majumder, N., & Gelbukh, A. (2017). Multimodal sentiment analysis: A survey
and analysis of recent work. ACM Computing Surveys (CSUR), 50(2), 1-37.
https://doi.org/10.1145/3053826
Sharma, N., & Singh, S. (2020). Data security and privacy issues in health care applications: A review.
International Journal of Computer Applications, 176(3), 1-8.
https://doi.org/10.5120/ijca2020919056
Larkin, G. L., & Beautrais, A. L. (2017). Suicide prevention and the role of emergency services: A global
perspective. Annals of Emergency Medicine, 69(4), 427-430.
https://doi.org/10.1016/j.annemergmed.2016.10.011
"Pinecone: A Vector Database for Machine Learning." Pinecone, https://www.pinecone.io/docs/.
Accessed November 2024.
Gupta, S., & Zhang, H. (2019). Artificial intelligence in content moderation: A survey. Proceedings of
the International Conference on Machine Learning and Data Engineering, 127-134.
https://doi.org/10.1145/3330759.3330784
Andersson, G., & Titov, N. (2014). Advantages and limitations of internet interventions for common
mental disorders. World Psychiatry, 13(3), 252-253. https://doi.org/10.1002/wps.20156