AI‑Powered Supply Chain Transparency for Counterfeit Detection (Demo)

This is a one‑day demo you can show live. It has three parts:

1. Product Authenticity Check

   • Upload a product photo. The app compares its perceptual hash to your trusted catalog images and estimates similarity.
   • Enter or scan a serial/QR string. The app runs format + Luhn‑style checks and validates against an allow‑list.

2. Transaction/Invoice Anomaly Detection

   • Upload a CSV (or use the sample). The app trains an IsolationForest and flags suspicious rows with scores.

3. Simple Supplier Risk View

   • Aggregates anomalies by supplier to produce a quick risk score and a bar chart you can show.

This demo avoids cloud dependencies so it runs completely offline.

---

Quickstart

# 1) (optional) create a venv
python -m venv venv && source venv/bin/activate   # on Windows: venv\Scripts\activate

# 2) install requirements
pip install -r requirements.txt

# 3) add 3–10 **trusted** product images to data/catalog/
#    (real product photos from your brand). Filenames become product IDs.
#    If you don't have images yet, the app still runs with a low-confidence warning.

# 4) run the app
streamlit run app.py

Then open the local URL Streamlit shows (usually http://localhost:8501).

---

Project Structure

ai_supplychain_counterfeit_demo/
├── app.py
├── requirements.txt
├── README.md
├── utils/
│   ├── image_match.py
│   ├── serial_check.py
│   └── anomaly.py
├── data/
│   └── catalog/        # put trusted product images here (jpg/png)
└── sample_data/
    └── sample_transactions.csv

---

How the Demo Works (talk track)

• Image match: computes a perceptual hash (pHash) for the uploaded image and compares to your trusted catalog with Hamming distance. Displays best match & similarity %. Below a threshold → possible counterfeit.
• Serial check: validates with regex pattern + Luhn‑like checksum; optional allowlist matches.
• Anomaly detection: trains IsolationForest on numeric features (amount, unit price, quantity, lead time). Rows with high anomaly scores are highlighted.
• Supplier risk: counts anomalies per supplier and normalizes into a 0–100 score you can explain to stakeholders.

---

Notes

• This is a reference demo, not production. For a real system:
  • Replace pHash with a fine‑tuned CNN/transformer or CLIP embedding search.
  • Back serial checks with a secure server & cryptographic signatures.
  • Use feature stores, lineage & audit trails, and a proper graph view.
• The thresholds are configurable in the UI.