A Multimodel Heathcare chatbot

🌱 I’m worked on  A Multimodel Healthcare Chatbot Using Deep Learning with Agentic Retrieval-Augmented Generation for Personalized Human-Computer Interaction for better medical queries.

This project focuses on developing a real-time, intelligent healthcare assistant powered by Agentic RAG (Retrieval-Augmented Generation) and the LangChain framework. Designed to provide multimodal support—text, voice, and document-based inputs—it aims to assist medical personnel and individuals in remote or defense settings. The system ensures timely interventions, supports diagnosis, and enables efficient access to critical medical knowledge, making it especially valuable in underserved or resource-constrained environments.

Artificial intelligence is rapidly transforming how we access healthcare, with AI-powered chatbots emerging as critical tools for providing immediate medical assistance. However, many traditional chatbots that rely solely on Large Language Models (LLMs) still have major shortcomings,12 traditional chatbot systems relying solely on Large Language Models (LLMs) face significant challenges, including outdated medical knowledge, They often provide outdated or inaccurate medical information, sometimes even making up facts (“hallucinations”), and they can raise serious privacy concerns when sensitive health data is sent to the cloud. This project presents HealthGenie, an intelligent medical chatbot that overcomes these limitations through an innovative integration of Retrieval-Augmented Generation (RAG) and Natural Language Processing (NLP) techniques.

 HealthGenie leverages a local LLM (Qwen2.5-7B) for secure, offline processing and combines it with a RAG pipeline built using LangChain to retrieve real-time medical information from curated sources (TATA 1mg, E-Raktkosh). The system employs spaCy-based intent detection to accurately classify user queries into three key domains: 

(1) symptom analysis and health advice,

 (2) medicine-related information (dosage, side effects, alternatives), and

 (3) blood bank location services. 

Key innovations include:  A privacy-preserving architecture that processes all data locally without external API calls  Structured response generation that provides diagnosis suggestions, actionable health advice, and verified resource links  Accurate entity recognition for medical terms, drug names, and geographic locations Evaluation results demonstrate that HealthGenie achieves 89% accuracy in symptom identification and responds within <2-3 seconds while completely eliminating data privacy risks associated with cloud based solutions. Compared to conventional LLM chatbots, HealthGenie reduces hallucination by 42% through context grounding via RAG. 

This work contributes a scalable framework for developing domain-specific, privacy-aware AI assistants in healthcare. Future extensions may incorporate multilingual support, voice interfaces, and real-time hospital system integrations. The complete implementation, including the Flask-based web interface and curated medical knowledge base, is made publicly available to support further research in applied medical AI.