Skip to content

AI Agent Demo

AortaCFD Agent turns a free-text clinical referral into a validated, literature-grounded patient-specific CFD configuration — fully automated, with citations for every parameter choice.

Open Interactive Demo GitHub

Interactive Demo

Try it directly — click Run Pipeline inside the app.

Not loading?

Open the Streamlit app directly. The demo renders the pre-computed BPM120 case (an 18-month-old paediatric coarctation) across the five agent stages.

Pipeline at a glance

Stage Agent LLM? What it does
1 IntakeAgent Yes Extracts structured patient profile from free-text
2 LiteratureAgent RAG Cites a 108-paper corpus for every parameter choice
3 ConfigAgent No Deterministically patches a template config
4 ExecutionAgent No Runs OpenFOAM (mesh · solver · post-processing)
5 ResultsAgent Yes Writes a grounded clinical summary from CFD output

Stages 3–4 are deterministic — once the intake and literature choices are locked, the config + simulation are fully reproducible.

Run it yourself

docker run --rm aortacfd-agent run \
  --case /app/external/aortacfd-app/cases_input/BPM120 \
  --clinical-text "18 month old male, aortic coarctation post-angioplasty, HR 120, BP 100/55" \
  --output /app/output/demo \
  --backend fake

git clone --recurse-submodules https://github.com/JieWangnk/aortacfd-agent.git
cd aortacfd-agent
pip install -e '.[anthropic]'
aortacfd-agent run \
  --case external/aortacfd-app/cases_input/BPM120 \
  --clinical-text "18 month old male, aortic coarctation post-angioplasty, HR 120, BP 100/55" \
  --output output/demo \
  --backend anthropic

aws batch submit-job \
  --job-name "BPM120-demo" \
  --job-queue aortacfd-queue-v2 \
  --job-definition aortacfd-cfd-run:3 \
  --container-overrides '{
    "command": ["run", "--case", "/app/external/aortacfd-app/cases_input/BPM120",
      "--clinical-text", "18-mo coarctation", "--output", "/app/output/demo",
      "--backend", "fake"]
  }'

Architecture

Clinical Text
    ▼
aortacfd-agent (LLM layer)
  ├── IntakeAgent      (Claude)
  ├── LiteratureAgent  (RAG, 108-paper corpus)
  └── ConfigAgent      (deterministic patching)
    ▼
AortaCFD-app (CFD core)
  ├── snappyHexMesh
  ├── Windkessel BCs
  ├── foamRun / PIMPLE
  └── TAWSS / OSI / RRT
    ▼
Clinical Summary

Provider-agnostic backends: Anthropic · OpenAI · Ollama · vLLM · OpenAI-compatible · fake (offline).

For the full bibliography, see References.

Found an issue or have a suggestion for this page?

Open a GitHub issue