Add clinical triage pipeline notebook for NMP v2

nemo/data-flywheel/e2e-llm-evaluation/README.md

@@ -0,0 +1,79 @@
+# Emergency Triage LLM Pipeline
+
+Synthetic data generation and model evaluation for ESI triage classification — powered by NeMo Microservices Platform.
+
+## Notebooks
+
+| Notebook                         | Description                                                   |
+|----------------------------------|---------------------------------------------------------------|
+| `clinical_triage_pipeline.ipynb` | Combined end-to-end NDD + Evaluator use case in one notebook |
+
+Run `clinical_triage_pipeline.ipynb` for the full pipeline in one place. It can be run locally with no GPU required.
+
+## Quick Start
+
+### 1. Create virtual environment
+
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+
+# Option A: with VPN / artifactory access
+pip install 'nemo-platform[data-designer]==2.0.0.dev1+nightly20260309' \
+  --index-url https://urm.nvidia.com/artifactory/api/pypi/nv-shared-pypi/simple
+
+# Option B: without VPN (e.g. on Brev) — download the wheel first via NGC CLI
+pip install 'nemo-platform-python-sdk_v2.0.0.dev1+nightly20260309/nemo_platform-2.0.0.dev1+nightly20260309-py3-none-any.whl[data-designer]'
+
+# Additional dependencies
+pip install -r requirements.txt
+python -m ipykernel install --user --name nmp --display-name "NMP (venv)"
+```
+
+### 2. Start NMP
+
+```bash
+nmp quickstart configure    # NGC key → NVIDIA Build inference → save
+nmp quickstart up --image nvcr.io/nvidian/nemo-llm/nmp-api:nightly-20260309
+nmp quickstart status       # Wait for health: ready
+```
+
+> **Important: Two different API keys are needed.**
+> - `nmp quickstart configure` requires the **NGC key** from the `nvidian/nemo-llm` org (https://org.ngc.nvidia.com/setup/api-key). This authenticates Docker image pulls.
+> - Step 3 below requires a **build.nvidia.com key** (https://build.nvidia.com). This authenticates LLM inference calls.
+>
+> If `configure` fails with `NGC login failed: unauthorized`, the NGC key is wrong — make sure you're in the `nvidian/nemo-llm` org when generating it.
+
+> **Troubleshooting:** If batch jobs (`data_designer.create()`) get stuck at `created`, verify the jobs controller started by checking `nmp quickstart logs` for `Backend registry initialized with: docker`. If missing, reinstall with the exact nightly version: `pip install 'nemo-platform[data-designer]==2.0.0.dev1+nightly20260309'` and do a clean `nmp quickstart destroy` + `configure` + `up`.
+
+### 3. Register build.nvidia.com provider
+
+Get your API key at [build.nvidia.com](https://build.nvidia.com) (separate from the NGC key used in step 2).
+
+```bash
+curl -s -X POST http://localhost:8080/apis/secrets/v2/workspaces/default/secrets \
+  -H "Content-Type: application/json" \
+  -d '{"name": "nvidia-build-api-key", "data": "<YOUR_BUILD_API_KEY>"}'
+
+curl -s -X POST http://localhost:8080/apis/models/v2/workspaces/default/providers \
+  -H "Content-Type: application/json" \
+  -d '{"name": "nvidiabuild", "host_url": "https://integrate.api.nvidia.com/v1", "api_key_secret_name": "nvidia-build-api-key"}'
+```
+
+### 4. Run
+
+Open the notebook in Jupyter or Cursor. Select the `NMP (venv)` kernel, then run `clinical_triage_pipeline.ipynb` for the full end-to-end flow.
+
+## Pipeline
+
+<img src="./triage_pipeline_diagram.png" alt="Pipeline Diagram" width="700" style="max-width:700px;"/>
+
+## NMP Commands
+
+```bash
+nmp quickstart status     # Check cluster health
+nmp quickstart logs       # View logs
+nmp quickstart down       # Stop cluster
+nmp quickstart destroy    # Stop and remove all data
+nmp quickstart doctor     # Diagnose issues
+```