🔥 Infherno

Infherno is an end-to-end agent that transforms unstructured clinical notes into structured FHIR (Fast Healthcare Interoperability Resources) format. It automates the parsing and mapping of free-text medical documentation into standardized FHIR resources, enabling interoperability across healthcare systems.

This repository contains the resources to the EACL 2026 System Demo paper Infherno: End-to-end Agent-based FHIR Resource Synthesis from Free-form Clinical Notes: https://aclanthology.org/2026.eacl-demo.13/

Built on Hugging Face’s SmolAgents library, Infherno supports multi-step reasoning, tool use, and modular extensibility for complex clinical information extraction.

Infherno also provides ontology support for SNOMED CT and HL7 ValueSets using Retrieval-Augmented Generation (RAG). This allows the agent to ground extracted medical concepts in standardized terminologies, ensuring semantic consistency and accurate coding in line with clinical data standards.

Live Demo

Gradio Demo Video: See the clip here.
Public Demo: A public demo is available at https://infherno.misit-augsburg.de/. It runs with an internal Snowstorm instance.

Our Gradio demo is accessible via Hugging Face Spaces. Although a Hugging Face Spaces instance was successfully running, we would like to ask you to use the other public demo above. Use of Local Models: For optimal performance and reliable results, it is recommended to use a strong commercial LLM like Gemini Pro 2.5, which was the model used in the experiments described in this paper. While local models were explored to a limited extent, their performance was observed to be substantially less reliable, and they were not used for the main evaluations. Also due to resource and context limitations with open-source models, we recommend launching Infherno locally with a proprietary model via API.

Run Infherno locally

Note: Expect most local models to yield substantially inferior results, as outlined in the previous section.

Install the dependencies first.

python3 -m venv env
source env/bin/activate

python3 -m pip install -r requirements.txt

Run the Infherno agent as follows:

# Define self-hosted Snowstorm instance
export SNOWSTORM_URL="http://<SNOMED-Instance>"

# Set Ollama endpoint
export OLLAMA_ENDPOINT="http://127.0.0.1:11434"

# Define custom open-weights model from Ollama to be used.
# MAKE SURE THAT THE MODEL IS ALREADY PULLED!
cat > local_config.py <<EOF
MODEL_ID = "ollama_chat/deepseek-r1:32b"
EOF

# Run the agent with dummy data
PYTHONPATH=. python3 infherno/smol_fhiragent.py

# Check the results in the logs:
cat logs/*.log

---

Citing Infherno

Infherno: End-to-end Agent-based FHIR Resource Synthesis from Free-form Clinical Notes
Johann Frei, Nils Feldhus, Lisa Raithel, Roland Roller, Alexander Meyer, Frank Kramer

ACL Anthology URL: https://aclanthology.org/2026.eacl-demo.13/

BibTex

@inproceedings{frei-etal-2026-infherno,
    title = "Infherno: End-to-end Agent-based {FHIR} Resource Synthesis from Free-form Clinical Notes",
    author = "Frei, Johann  and
      Feldhus, Nils  and
      Raithel, Lisa  and
      Roller, Roland  and
      Meyer, Alexander  and
      Kramer, Frank",
    editor = "Croce, Danilo  and
      Leidner, Jochen  and
      Moosavi, Nafise Sadat",
    booktitle = "Proceedings of the 19th Conference of the {E}uropean Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 3: System Demonstrations)",
    month = mar,
    year = "2026",
    address = "Rabat, Marocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.eacl-demo.13/",
    doi = "10.18653/v1/2026.eacl-demo.13",
    pages = "163--174",
    ISBN = "979-8-89176-382-1",
    abstract = "For clinical data integration and healthcare services, the HL7 FHIR standard has established itself as a desirable format for interoperability between complex health data. Previous attempts at automating the translation from free-form clinical notes into structured FHIR resources address narrowly defined tasks and rely on modular approaches or LLMs with instruction tuning and constrained decoding. As those solutions frequently suffer from limited generalizability and structural inconformity, we propose an end-to-end framework powered by LLM agents, code execution, and healthcare terminology database tools to address these issues. Our solution, called Infherno, is designed to adhere to the FHIR document schema and competes well with a human baseline in predicting FHIR resources from unstructured text. The implementation features a front end for custom and synthetic data and both local and proprietary models, supporting clinical data integration processes and interoperability across institutions. Gemini 2.5-Pro excels in our evaluation on synthetic and clinical datasets, yet ambiguity and feasibility of collecting ground-truth data remain open problems."
}

ArXiv

ArXiv URL: https://arxiv.org/abs/2507.12261

@article{frei-2025-infherno,
    title = {Infherno: End-to-end Agent-based FHIR Resource Synthesis from Free-form Clinical Notes}, 
    author = {Johann Frei and Nils Feldhus and Lisa Raithel and Roland Roller and Alexander Meyer and Frank Kramer},
    year = {2025},
    volume = {abs/2507.12261},
    journal = {arXiv},
    primaryClass = {cs.CL},
    url = {https://arxiv.org/abs/2507.12261}, 
}