Healthmate: An AI Chatbot for Symptom-Based Healthcare Consultation

Abstract

ABSTRACT In modern healthcare systems, preliminary diagnostic guidance is often constrained by high consultation costs, long clinic waiting times, and limited access to healthcare professionals in underserved regions. Consequently, individuals frequently turn to online search engines, leading to self-diagnosis errors and anxiety caused by unstructured misinformation. Traditional automated symptom checkers rely heavily on rigid rule-based systems or basic keyword matching, failing to capture semantic nuances, user phrasing variations, and clinical context. To address these limitations, this paper presents HealthMate, an intelligent, explainable AI chatbot framework designed for preliminary healthcare consultation. The proposed system utilizes Natural Language Processing (NLP) and Sentence-BERT (all-MiniLM-L6-v2) transformers to convert natural language symptom descriptions into 384-dimensional dense semantic embeddings. High-speed similarity search is conducted against a structured medical knowledge base using Facebook AI Similarity Search (FAISS). Furthermore, a novel hybrid scoring mechanism is developed, combining dense semantic similarity with explicit token-level symptom overlap coverage to re-rank candidate conditions accurately. To eliminate "black-box" outputs, a Retrieval-Augmented Generation (RAG) architecture powered by a local LLaMA-3 language model synthesizes transparent diagnostic reasoning, precautionary guidance, and specialist advice. Implemented using Python and Streamlit, HealthMate demonstrates rapid retrieval, robust natural language comprehension, and clear explainability without replacing professional medical diagnosis. Keywords: Medical Chatbot, Natural Language Processing, Sentence Transformers, FAISS, Hybrid Scoring, Retrieval-Augmented Generation (RAG), Healthcare AI, Streamlit.