📦 Warehouse Routing Intelligence

A single-file, zero-dependency web app that predicts which warehouse a parcel belongs to — from a barcode scan, a tracking code, or a camera photo of a shipping label. Runs entirely in the browser; no backend, no installation.

---

✨ Features

Prediction engine

• Exact code lookup — instant match against a trained dataset of known tracking IDs
• Prefix / pattern matching — rule-based engine that reads code structure (e.g. BE…KESS → Kess Berlin → POZ2)
• Merchant keyword extraction — scans the tracking code itself for embedded merchant tokens (e.g. KESS, JACKS) and maps them to the correct warehouse
• Merchant hint lookup — type or scan a merchant name for direct routing
• Base-rate fallback — when no signal is found, uses historical warehouse distribution as a prior
• All layers are combined with confidence scoring; the highest-confidence signal wins

Input methods

Method	Desktop	Mobile
USB / Bluetooth HID barcode scanner	✅	—
Manual keyboard entry	✅	✅
Camera barcode scan (BarcodeDetector API)	✅	✅
Camera OCR label scan (Tesseract.js)	✅	✅

Multi-device sync

• Create a session on the desktop to generate a 6-character code
• Scan the QR code shown in the sidebar to connect a phone instantly
• Both devices share the same prediction results in real time via PeerJS (WebRTC)

Dataset upload (Settings panel)

Upload two CSV files to replace the built-in dataset at runtime — no page reload needed:

File	Required columns
Merchant lookup	Merchant, Warehouse
Tracking ID dataset	TrackingID, Merchant, Warehouse

Click 🔄 Train new model after uploading to rebuild the prediction engine with your data.

Dashboard

• Live warehouse split bars (BER3 / POZ1 / POZ2 / other)
• Merchant breakdown table — predicted merchant name, count, and percentage
• Recent scans log with warehouse badge, confidence, and timestamp
• Session stats counter strip (mobile-friendly)

UX

• 🌙 / ☀️ Dark / light theme toggle
• ⚙️ Settings panel (desktop only) for dataset management
• Fully responsive — switches to a focused mobile scan view automatically
• QWERTZ → QWERTY remapping for HID scanners on German keyboard layouts — scans (Y)#13796 correctly even when the OS layout produces )Z=§13796
• Strips leading (Y) prefix automatically before prediction

---

🚀 Getting started

No build step. Just open the file.

# Clone and open
git clone https://github.com/YOUR_USERNAME/warehouse-routing-intelligence.git
open warehouse_predictor_app.html

Or host it anywhere static — GitHub Pages, Netlify, a local file server.

With your own data

1. Open the app and click ⚙️ (bottom-left of the sidebar)
2. Upload your Merchant lookup CSV (Merchant, Warehouse)
3. Upload your Tracking ID dataset CSV (TrackingID, Merchant, Warehouse)
4. Click 🔄 Train new model
5. Start scanning

Both comma (,) and semicolon (;) delimiters are supported. Column headers are case-insensitive.

---

📁 File structure

warehouse_predictor_app.html   ← entire app (HTML + CSS + JS, single file)
README.md

---

🧠 Prediction logic

Input code
    │
    ├─ Exact match in EXACT{}?           → warehouse (100% confidence)
    │
    ├─ Prefix / pattern rule match?      → warehouse + confidence score
    │
    ├─ Merchant hint provided?
    │     └─ MERCHANT_PURE / MIXED lookup → warehouse + confidence
    │
    ├─ Merchant token found in code?
    │     └─ Substring scan of code body  → warehouse + confidence (85%)
    │
    └─ None of the above?                → base-rate warehouse (BER3 ~73%)

Multiple signals → highest confidence wins
Ties → merchant signal preferred over base rate

---

📦 Dependencies (CDN, no install)

Library	Purpose
PeerJS	WebRTC multi-device sync
QRCode.js	QR code generation for session joining
Tesseract.js	OCR for camera label scanning
Syne + JetBrains Mono	Typography

All loaded from CDN. The app works offline after first load (fonts and Tesseract model cached by the browser).

---

🔒 Privacy

All processing happens in your browser. No tracking codes, merchant names, or session data are sent to any server. PeerJS uses a public signalling server only to establish the WebRTC connection — the scan data itself flows peer-to-peer.

---

🛠 Development notes

• Single HTML file — edit CSS in <style>, logic in <script>
• Prediction data lives in const EXACT, const MERCHANT_PURE, const MERCHANT_MIXED near the top of the script — replace with your own or load via CSV upload
• To add a new warehouse, add entries to those objects and update the warehouse stat tiles in the HTML

---

📄 License

MIT — use freely, attribution appreciated.