SmartPackageBox

An integrated, fully automated IoT home system designed to intercept and secure delivered packages without human intervention. Built natively as a technological thesis project, the architecture merges physical relay controls on edge devices with high-level Vision Transformer (ViT) classification models and cross-platform companion software.

---

📖 Official Documentation

For deep technical insights regarding the structural engineering, software data-flows, AI data-labeling scripts, and Avalonia Application logic, refer strictly to the official GitHub Pages deployment:

🌐 Read the Full Documentation Portfolio Here

Core Documentation Modules

• Hardware Specifications: Schematics on the Raspberry Pi Zero 2W, Somfy LT50 Motor, and opto-isolated 2-channel 230V relays.
• Software Architecture: MVVM patterns, .NET 9 internal structures, buffer flushing via memory arrays, and HTTP API design.
• Companion Application (Avalonia UI): Breakdown of the cross-platform telemetry application parsing WebAssembly, Android APKs, and external settings.
• Vision Transformer AI: Details on the Batch-Annotate.sh data labeling parameters, ViT classification strategies, Base64 network tunneling payloads.

---

System Architecture

The physical logic executes from a central localized device (Raspberry Pi Zero 2W) which commands automated mechanisms upon reading valid inferential data.

1. Detect: The system actively surveys the internal physical volume using an integrated Logitech C270 array.
2. Classify: Image bytes are extracted and converted to Base64, then tunneled toward an offline inference homelab executing a Binary Vision Transformer model.
3. Actuate: Upon receiving a valid "Package" boolean, the MainController engages the 230V AC Motor mechanical arrays, permanently sealing the compartment.
4. Notify: Pushes remote images via ntfy.sh and actively serves telemetry to the Avalonia Client dashboard.

AI Dataset & Repository

The core tensor parameters are fully accessible. The ViT model structure was constructed over roughly 10,000 unique annotated image sets validating interior light curves, cardboard structures, and ambient occlusions.

• Roboflow Dataset: Package Detection Universe