Industrial
Architecture

Stack: Supabase, Gemini 1.5 Pro, PWA with IndexedDB Secure Sync.

Forensic Data Flow

graph TD A[Secure Capture] -->|Pixels + Metadata| B[Offline IndexedDB] B -->|Synchronization| C[Supabase Storage] C -->|Webhook| D[Gemini Vision Engine] D -->|Delta Analysis| E{Red Alert?} E -->|Yes| F[Push Manager / Lockout] E -->|No| G[Operational Approval]

Computer Vision

We implement Gemini 1.5 Pro for visual disparity analysis. The engine processes comparisons for:

- Tire tread patterns
- Numbered security seal integrity
- OCR on analog meters (Rotogauge)

Hardened Camera

Capture occurs exclusively in the live video buffer.

- Gallery Lock (Input File Disabled)
- Watermark Burn-in (Canvas Layer)
- GPS Cross-check with IP Geolocation

Module: Fixed Gate Audit (Coming Soon)

System extension for total automation at terminal gates using fixed computer vision.

Fixed Hardware

Integration with high-framerate IP cameras (RTSP) positioned at ground level for lateral capture.

Flow AI

Segmentation algorithms to identify each individual tire while the trailer is in motion.

AI Extraction Protocol (System Prompt)

Role: Forensic Tire Expert

ACTÚA COMO UN PERITO FORENSE DE NEUMÁTICOS. Analiza esta ÚNICA imagen de una llanta. OBJETIVO: Extraer TODA la información de identidad visible. Busca Específicamente: 1. MARCA (Brand) 2. MODELO (Model) 3. NÚMERO DE SERIE / DOT (Serial Number) - ¡CRÍTICO! Busca patrón de 4-13 caracteres. 4. DESGASTE ESTIMADO (mm). 5. ESTADO DEL RIN (Rim Condition) - "Bueno", "Doblado", "Oxidado", "Quebrado". RESPONDE SOLO JSON: { "marca": "string o null", "modelo": "string o null", "serial_number": "string o null", "profundidad_huella_mm": number (0-20), "estado_rin": "string o null" }

Reference: src/services/ai/gemini.ts v2.5

Advanced Intelligence Stack [LIVE]

1. ALPR (License Plate)

Local edge processing for instant vehicle identification in the browser.

- Engine: Tesseract.js OCR
- Latency: < 200ms (Client-Side)
- Auto-fill in Inventory

2. AI Biometrics

Facial micro-expression analysis for occupational health monitoring.

- Engine: Face-API.js (TensorFlow)
- Detection: Fatigue, Stress (Anger/Fear)
- HUD: Real-time Visualization

3. Critical SOS Protocol

Immediate response system for high-risk incidents (Assault / Confirmed Microsleep).

Visual Alert: Global Toast in Dispatcher Dashboard
Webhook (Edge): Auto Send WhatsApp/Telegram

Hardware Infrastructure (Phase 6)

1. Site Edge (Depot)

Heavy local processing for 24/7 zero-latency analysis.

» Server: Industrial Mini PC (NVIDIA Jetson)
» Cameras: IP CCTV via PoE Ethernet connection
» Function: RTSP Ingest -> AI Analysis -> Alert-Only Sync

2. Mobile Edge (On-Road)

Lightweight monitoring using existing devices.

» Option A: SIMSA App on Phone (Local Face Detection)
» Option B: 4G AI Dashcam (Clips on demand)
» Function: In-Cab Audio Alert + Push to Control Tower

Resilience Protocol (Backup)

Offline-First architecture designed for contingencies:

Level 1: App (IndexedDB) If network fails, data is encrypted locally on device and auto-syncs upon reconnection.
Level 2: Server (Cold Storage) Physical video backup on local HDD (30-day loop) independent of cloud status.

Restricted Access