Abstract:
Despite stigma, high costs, and resource scarcity, mental health challenges are
increasingly afflicting people in all generations, but support on time, in personal and
affordable manners, is still not easily reached. For this reason, Artificial Intelligence
is taking on a nurturing role as a means of augmenting emotional support and scale
Generative AI-Based Mental
delivery of therapeutic interventions. In this project, a
and Emotional Health Support System is presented, which is built upon a modular
agent-based architecture to simulate a CBT inspired support. The system is built using
the Gemini API and LangChain, integrating nine specialized tools such as a Retrieval
Tool (Pinecone), Mood Analysis, Thought Reframing, Crisis Detection, Reflective
Listening and a Web Search component (Tavily). The tool set consists of each tool as
central executor that dynamically
agent coordinated through a
an autonomous
determine user input and select the appropriate tool for instance and intent. The prompt
engineering is based on modular template design with structured instructions, persona
and few shot learning to guarantee tone, safety and coherence. Retrieval
in relation to scientifically validated
settings
Augmented Generation (RAG) puts responses
factual accuracy. It has a private and seamless
mental health literature to increase
interaction through a Streamlit interface. Crisis detection and topic filtering provide
ethical safeguards to protect the user. Users and psychologists had posmve feedback
on the system as a potential reliable and emotionally intelligent digital mental health
companion.
