Llm-based Medical Ai Chatbot Using Llama For Healthcare Assistance

Abstract

Access To Quality Healthcare Remains A Critical Global Challenge. Approximately 3.5 Billion People Lack Access To Basic Healthcare Services. This Paper Presents The Design And Development Of A Full-stack Document Intelligence Application Powered By Architecture For Intelligent Healthcare Query Support. The System Allows Users To Upload Documents (PDF, DOCX, TXT), Which Are Semantically Chunked, Embedded Using Google Generative AI Embedding, And Stored In A FAISS Vector Store. Upon Receiving A User Query, Relevant Documents Are Retrieved From FAISS And Passed As Context To The Gemini-2.0-Flash Model, Which Generates Accurate, Hallucination-reduced Responses Grounded In Verified Medical Literature. The Proposed Architecture Incorporates A Curated Medical Knowledge Base Scraped From WebMD (1,613 Health Topics, 27,744 Chunks), Chain-of-thought Prompt Engineering, Session-aware Chat History Management, And A Multimodal Image-based Retrieval Pipeline Using Gemini Flash 2.0. The System Integrates Textual Prompt Analysis And Medical Image Interpretation, Supporting Preliminary Medical Diagnosis For Underserved Communities. Experimental Evaluation Across 30 Conversation Sessions And 119 Medical Queries Demonstrates An Average Response Time Of 3.6 Seconds, Relevance 4.8/5, Fluency 4.9/5, And User Safety 4.2/5, Substantially Outperforming Standalone LLM Querying.

Keywords