Skip to content

Cookbook - Streamlit stakeholder demo

You're presenting arche-core to a stakeholder - a procurement officer at a Lagos health-tech buyer, a programme manager at an African NLP funder, a regulator's deputy who wants to see what "statute-grounded PII detection" means in practice. They have ten minutes. They are not going to read the README. They are going to look at a screen.

Before arche-core: you screenshot terminal output and hope it lands. The screenshots show JSON dumps. The stakeholder politely asks how this is different from Presidio.

With arche-core: a fifty-line Streamlit app turns the same Pipeline.process() call into something that lands in seconds. Paste text, pick a jurisdiction, see detections with their statute citations, see the redacted output, copy a signed JWS receipt.

# app.py
import streamlit as st
from arche import Pipeline
from arche.sign import SignWorkflow, generate_keypair

st.set_page_config(page_title="arche-core demo", layout="wide")
st.title("African PII detection - with the law it operates under built in")

JURISDICTIONS = {
    "Nigeria - NDPA-2023": "NG",
    "South Africa - POPIA": "ZA",
    "Kenya - Kenya DPA": "KE",
    "Ghana - Ghana DPA": "GH",
}

col_left, col_right = st.columns([1, 1])

with col_left:
    label = st.selectbox("Jurisdiction", list(JURISDICTIONS.keys()))
    jurisdiction = JURISDICTIONS[label]
    sample = st.text_area(
        "Paste text to scan (synthetic only - never real PII)",
        height=240,
        value=(
            "Customer Adesola Okonkwo registered with NIN 12345678901 "
            "and BVN 22156789012. Contact phone 0803 555 7890. "
            "Company: RC 245678."
        ),
    )
    run = st.button("Detect + apply policy", type="primary")

if run and sample.strip():
    pipeline = Pipeline(jurisdiction=jurisdiction, tokenize_salt="demo_2026")
    result = pipeline.process(sample)

    with col_right:
        st.subheader("Detections")
        st.dataframe(
            [
                {
                    "Category": d.category,
                    "Tier": d.sensitivity_tier.value,
                    "Citation": d.regulatory_citation,
                    "Text": d.text,
                }
                for d in result.detections
            ],
            hide_index=True,
            use_container_width=True,
        )

        st.subheader("Redacted text (safe to share)")
        st.code(result.redacted_text, language="text")

        st.subheader("Policy outcomes - the rule that fired")
        st.dataframe(
            [
                {
                    "Category": o.category,
                    "Action": o.action,
                    "Statute reference": o.statute_reference,
                }
                for o in result.policy_outcomes
            ],
            hide_index=True,
            use_container_width=True,
        )

        # Optional: signed receipt the stakeholder can copy + verify
        if st.checkbox("Generate signed JWS receipt"):
            demo_key = generate_keypair()
            signed = SignWorkflow(jurisdiction=jurisdiction).sign(
                sample, demo_key, purpose="stakeholder_demo"
            )
            st.subheader("Signed JWS envelope")
            st.code(signed, language="text")
            st.caption(
                f"Issuer did:key: {demo_key.did_key[:40]}... "
                "Verify offline with arche.sign.VerifyExtractWorkflow."
            )

st.divider()
st.caption(
    "Demo runs locally. No data leaves your machine. "
    "Source: github.com/unpatterned-labs/arche. License: Apache 2.0. "
    "Status: pre-beta (development) - not for production use."
)

Run it:

pip install arche-core streamlit
streamlit run app.py

The browser opens at http://localhost:8501. Paste any of the four preloaded samples (NG / ZA / KE / GH). Click Detect + apply policy. The detections show up with sensitivity tier and the actual statute section the policy mapping cites - not as a JSON blob but as a table the stakeholder can read.

What the stakeholder takes away

When the stakeholder watches the demo and sees NDPA-2023 s.30, NIMC Act s.27 rendered next to the NIN they pasted, they internalise "oh, this is the legal layer wrapped around the detection layer." That sentence is the pitch. It lands in a second when they see it; it takes a paragraph when you describe it.

When they switch the jurisdiction dropdown from Nigeria to South Africa and watch the citations change to POPIA, they internalise "one tool, four jurisdictions." That's another second.

When they tick Generate signed JWS receipt and see the envelope, they internalise "and the audit trail is cryptographic." Third second.

You are now three seconds in and they have understood the product shape. Use the remaining time to ask what they are trying to build.

Deployment options

Where When
Local laptop Internal demos, evaluation, sales calls
Streamlit Community Cloud Public live demo (e.g. demo.unpatterned.org) - free tier is fine for stakeholder demos
Docker container behind your reverse proxy When the stakeholder wants to evaluate on their own data inside their perimeter

The Streamlit Community Cloud deployment of this script (with four preloaded samples) is exactly what's at demo.unpatterned.org. Source: demo/app.py.

See also