Mental Healthcare Chatbot Based on Custom Diagnosis Documents Using a Quantized Large Language Model
Ayush Kumar, Sanidhya Sharma, Shreyansh Gupta, Dharmendra Kumar
Abstract
This research presents a novel retrieval-based question-answering (QA) framework utilizing LangChain's (version 0.1.6) modular architecture and Chainlit's (version 0.7.700) conversational interface. Our system efficiently handles PDF and directory documents, generates sentence embeddings with HuggingFace's pre-trained model, stores vectors in FAISS for fast search, employs the powerful CTransformers (version 0.2.27) with Llama-2-7B-Chat-GGUF model, and guides it with a custom prompt template for accurate and factual responses. The integrated Chainlit interface facilitates user interaction, demonstrating the framework's potential for knowledge-intensive domains like medical chatbots.