docs/user_guide.md

User guide

Prerequisites

• Python 3.10 or later
• A free Google AI Studio API key
• A Tavily API key (free tier available)
• A LangSmith API key (free tier available)

Installation

git clone https://github.com/chaeyoonyunakim/noteguard-agent.git
cd noteguard-agent

python -m venv .venv
.venv\Scripts\activate          # Windows
# source .venv/bin/activate     # macOS / Linux

pip install -e ".[dev]"

To also install the Streamlit de-id demo:

pip install -e ".[demo]"       # Streamlit interactive demo

Configuration

cp .env.example .env

Open .env and fill in your credentials:

GOOGLE_API_KEY=AIza...
TAVILY_API_KEY=tvly-...
LANGSMITH_API_KEY=ls__...
LANGSMITH_TRACING=true
# NOTEGUARD_MODEL=google_genai:gemini-2.5-flash   # optional override

Running the de-identification demo (no API keys needed)

python src/deid.py

Demonstrates the de-id core on a synthetic note — no network calls, no keys.

Running the interactive Streamlit demo (no API keys needed)

streamlit run streamlit_app.py

Lets you paste any text, click De-identify, and see the surrogate-token output alongside the vault contents and assert_clean result.

Downloading the dataset

python src/fetch_dataset.py

Downloads patients.csv, admissions.csv, and synthetic_clinical_notes.csv from the NHSEDataScience/synthetic_clinical_notes Hugging Face dataset into data/. Run once before starting the server to enable the note-picker and vault-based leakage metrics.

Running the clinician web UI (recommended for demos)

uvicorn app.api:app --reload --port 8000

Open http://localhost:8000.

1. Click Load note (top-right) to open the note picker, or paste your own note.
2. Click Generate (~20–30 s on first run; the model loads and de-identifies).
3. Use the segmented toggle to switch views without re-calling the API:
   • Clinician view — the original note with every redacted identifier highlighted in red.
   • What the AI sees — the de-identified note; real identifiers are replaced by [TYPE_N] surrogate chips (e.g. [PERSON_1], [NHS_1]).
4. The compact eDischarge card appears on the right, re-identified for the clinician.
5. The trust panel below shows whether de-identification was done correctly — and nothing about answer quality:
   • De-identification — PASS only when nothing un-redacted reached the model and every surrogate is reversible; FAIL otherwise.
   • Identifiers replaced — count of PII spans pseudonymised in this call.
   • Residual PII · model input — suspected un-redacted identifiers the model still saw (0 = clean). When > 0, the offending snippets are listed (e.g. name: Ethel Joanne Duffy). This catches free-text names the vault/NER passes missed — the case the old re-id-risk number was blind to.
   • Reversible — ✓ when every surrogate restores to a real value; ✗ lists the orphaned/unresolved tokens.
6. Click ← Edit note to reset and process a different note.

Running the LangGraph dev server

langgraph dev

Connect the Agent Chat UI and interact with the noteguard graph directly.

Running the LangSmith evaluations

python -m eval.run_eval

Needs LANGSMITH_API_KEY and LANGSMITH_TRACING=true. Runs two evaluators:

Evaluator	Target	What it measures
zero_phi_to_model	1.0	No known identifier appeared in any message seen by the model.
faithfulness	0.8+	Every clinical claim in the answer is supported by the source note.

Development

ruff check src agent app eval tests   # lint
ruff format src agent app eval tests  # format
pytest                                 # run the test suite
pytest --cov=src --cov-report=term    # with coverage

Loading a real patient vault

The de-id core can be seeded from the NHSEDataScience/synthetic_clinical_notes dataset (MIT licence, fully synthetic):

from src.deid import NoteGuard, load_known_from_csv

known = load_known_from_csv("data/patients.csv", "data/admissions.csv")
ng = NoteGuard(known=known)
result = ng.deidentify(raw_note)
ng.assert_clean(result.clean_text)

This builds the identifier vault from both structured tables — patient names and clinician names — so residual leakage is measured against ground truth.