AI System for Querying FHIR Data via Natural Language
FHIR NLP System - Devpost
Summary
This project built an interoperability connector using NLP and AI, allowing non-specialists to gain insights from medical data simply by asking questions in plain English. The system automatically converts user queries into optimized SQL and displays FHIR-compliant data visually.
Details
This AI-powered natural language query engine allows users without specialized knowledge to ask simple questions, such as 'Show me all patients with diabetes,' regarding complex healthcare data (FHIR R4). The core functionality is converting the natural language input into automatically optimized BigQuery SQL queries using LLMs like Gemini. This eliminates the need for users to understand complex FHIR schemas or write SQL. Technically, it uses a Python Flask API and a React frontend, querying the public FHIR Synthea dataset (17 resource types) hosted on Google BigQuery. Key development challenges included handling FHIR's deeply nested structures and implementing smart formatting to prevent raw object displays. This resulted in an end-to-end MVP that not only generates queries but also visualizes results in a readable format, making complex data accessible.
Original content copyright by respective publishers