FHIR-RAG-MED Interpretation System

Overview

The FHIR-RAG-MED Interpretation System is a backend module designed for interpreting FHIR Bundles in compliance with the SMART on FHIR specification. This system allows users to fetch FHIR data from a SMART-enabled FHIR server and process it for medical recommendations using either a cloud-based API or a local Llama model.

Configuration

The system is configured via config.py:

• OPENROUTERAI_API_KEY: API key for OpenRouter AI (Required if running bundle interpretation via cloud API).
• RUN_BUNDLE_INTERPRETATION_LOCAL: Determines how FHIR bundle interpretation is executed:
  • False (Default): Uses OpenRouter AI for interpretation.
  • True: Runs interpretation locally using Ollama.
    • Install Ollama.
    • Pull the model (if your machine has the necessary resources): ollama pull llama3.3:70b

Installation

1. Extract the repository.
2. Install dependencies:

   pip install -r requirements.txt

Running the System

The system can be run in two modes depending on the availability of the FHIR Bundle.

1. Running as a Backend Service

If the FHIR Bundle needs to be fetched first:

Using Docker

docker compose up

Using a Local Python Environment

backend.bat  # Windows
# OR
set PYTHONPATH=. && python src/app/app.py  # Manually running the backend

After starting the backend, use the following cURL requests to fetch and interpret the FHIR Bundle:

Fetch FHIR Data

curl 'http://localhost:8000/fetch-fhir-data'   -H 'Accept: application/json, text/plain, */*'   -H 'Content-Type: application/json'   --data-raw '{"access_token":"eyJhbGciO...","patient_id":"pat12","fhir_server":"https://kroniq.srdc.com.tr/fhir"}'

Interpret FHIR Data

curl 'http://localhost:8000/interpret_fhir'   -H 'Accept: application/json, text/plain, */*'   -H 'Content-Type: application/x-www-form-urlencoded'   --data-raw 'fhir_bundle='

2. Running with an Existing FHIR Bundle

If you already have a FHIR Bundle, you can run the interpretation directly via command line:

set PYTHONPATH=. && python src/core/recommendation/interpretation.py -b fhir_bundle_file.json

Notes

• Ensure the FHIR server is SMART-on-FHIR enabled.
• When using Docker, the service will be available at http://localhost:8000/.
• The system supports both remote API and local model-based interpretation for flexibility.