OPENMIND_AI-RAG PROJECT

This project implements a mental support chatbot using a Retrieval-Augmented Generation (RAG) framework. The chatbot is designed to provide compassionate, thoughtful, and context-aware responses by leveraging pre-existing datasets and integrating advanced natural language processing techniques.

---

Project Overview

The chatbot:

• Uses a csv file of psychological contexts and responses to generate insightful answers.
• Embeds data using SentenceTransformers and stores it in a Pinecone vector database.
• Retrieves relevant context dynamically during conversation.
• Fine-tunes an LLM model to act as a compassionate, friendly personal assistant.
• Processes queries to provide meaningful, concise, and empathetic responses.

---

Workflow

1. Data Preprocessing:
   • The data is taken from huggingface, removed the third column from it because of irrelevant responses.
   • Converted to csv using pandas.
   • Load the dataset using LangChain's CSVLoader.
   • Format the data into question-answer pairs.
2. Text Chunking:
   • Use LangChain's RecursiveCharacterTextSplitter for chunking large texts.
3. Embeddings:
   • Generate vector embeddings using SentenceTransformers (all-MiniLM-L6-v2).
4. Storage:
   • Store embeddings in Pinecone for fast and scalable retrieval.
5. Retrieval-Augmented Generation:
   • Retrieve context based on user queries.
   • Use LLaMA 3.3 for generating responses augmented with retrieved context.

---

Technologies Used

• Language Models: LLaMA 3.3
• Embeddings: SentenceTransformers (all-MiniLM-L6-v2)
• Database: Pinecone
• Frameworks: LangChain, Streamlit
• Programming Language: Python