BD Medicine AI: An AI-Powered Emergency Healthcare and Pharmaceutical Information Assistant Using Hybrid Retrieval-Augme

During earthquakes, floods, cyclones, and other disasters, access to reliable healthcare information becomes difficult as medical services are disrupted and healthcare professionals are overwhelmed. Patients may only have a handwritten prescription or the name of a medicine but lack information about its dosage, generic alternatives, side effects, or precautions. At the same time, people often struggle to locate nearby emergency shelters or obtain verified shelter information quickly. Existing information is scattered across multiple platforms, making it difficult for patients, caregivers, volunteers, and frontline responders to make timely decisions. A single platform that provides trusted pharmaceutical guidance alongside accessible emergency shelter information can significantly improve emergency preparedness and response during crisis situations.

BD Medicine AI is an AI-powered emergency healthcare assistant designed to improve access to reliable pharmaceutical information during disasters. The platform uses a hybrid Retrieval-Augmented Generation (RAG) framework with a localized knowledge base containing over 21,000 Bangladeshi medicines to answer questions about medicine usage, dosage, generic alternatives, indications, side effects, and precautions. Users can search by medicine name or ask questions in natural language to receive fast, context-aware responses. In addition to AI-powered medicine assistance, the platform includes an interactive Bangladesh map and a searchable emergency shelter directory built from official government data. Users can explore districts on the map, search for emergency shelters, and view shelter details in both map and table formats, making verified shelter information easier to access during emergencies. While the shelter information is currently presented as a searchable resource and is not yet integrated with the AI assistant, it complements the healthcare features by providing essential disaster-response information in a single platform. Together, these features help patients, caregivers, volunteers, and emergency responders make informed decisions when timely access to information matters most.

Languages Python 3.11 – Backend application development and AI integration. JavaScript (ES6+) – Frontend application logic. Frontend Frameworks React 18 – User interface development. Vite – Fast frontend build tool and development server. Backend Framework FastAPI – RESTful API development and communication between frontend and AI services. Database PostgreSQL – Stores medicine records, user data, and search history. ChromaDB – Vector database for semantic search in the RAG pipeline. AI Models Llama 4 Scout Vision (via Groq API) – Generates answers to medicine-related questions and extracts information from prescription images. all-MiniLM-L6-v2 (SentenceTransformers) – Creates text embeddings for semantic retrieval. Groq Whisper – Converts voice input into text. APIs & Cloud Services Groq API – Provides LLM inference and speech processing. Firebase Authentication – User authentication and account management. Cloudinary – Secure storage and delivery of uploaded prescription images. Mapping & Visualization MapLibre GL JS v6 – Displays Bangladesh districts and emergency shelter locations on an interactive map. Third-Party Data Sources (Attribution) Bangladesh Emergency Shelter Data Source: Department of Disaster Management (DDM), Ministry of Disaster Management and Relief, Government of Bangladesh. Website: https://www.ddm.gov.bd Bangladesh District GeoJSON (bd_adm2.json) Source: Department of Disaster Management (DDM), Government of Bangladesh. Medicine Database (21,000+ Bangladeshi Medicines) Compiled from publicly available Bangladeshi pharmaceutical information (including MedEx and other publicly accessible sources) and indexed for semantic retrieval.

Groq (Llama 4 Scout Vision) Used for Retrieval-Augmented Generation (RAG), medicine question answering, and prescription OCR. Groq Whisper Used for speech-to-text transcription of voice medicine queries. all-MiniLM-L6-v2 Used as a pre-trained embedding model for semantic medicine retrieval. Gemini Used for development assistance, debugging, and documentation. Antigravity IDE Used for code completion, development assistance.