Skip to main content

🏗 System Architecture

ChildSafeNet is a hybrid AI-powered internet safety platform combining real-time URL detection, AI inference, and browser-level enforcement.

ReactASP.NET CoreSQL ServerHybrid AIBrowser Extension

Platform Overview​

ChildSafeNet consists of five main layers:

🌐 Web Application​

Parent/Admin dashboard for configuration and monitoring.

⚙️ Backend API​

Central control layer handling authentication, scanning, and logging.

🗄 Database​

Stores users, policies, logs, and datasets.

🤖 AI Engine​

Hybrid ML model for URL classification.

🧩 Browser Extension​

Enforces blocking decisions in real time.

System Diagram​

System Architecture Diagram of ChildSafeNet

1. High-Level Architecture​

1
Frontend

React Web App​

Parent and admin interface used for configuration, monitoring, and system management.

2
Backend

ASP.NET Core API​

Central backend responsible for authentication, scanning endpoints, logging, and policy enforcement.

3
AI

Hybrid AI Engine​

Machine learning model used to detect malicious, adult, or suspicious content.

2. Core Components​

🌐 React Web (Parent/Admin UI)​

Responsibilities:

  • Authentication (Parent/Admin)
  • Protection mode configuration
  • Whitelist / Blacklist management
  • Scan log visualization
  • Dataset review (Admin)
  • Trigger model training jobs
  • Monitor model metrics

The web application communicates only with the Backend API via secured REST endpoints.


⚙️ ASP.NET Core API​

Responsibilities:

  • JWT authentication
  • URL scanning endpoint (/api/scan)
  • Parent settings management
  • Scan log persistence
  • Dataset collection
  • Training orchestration
  • Model registry management

The API acts as the central control layer connecting Extension, AI Engine, and Database.


🗄 SQL Server Database​

Main tables:

  • Users
  • UserSettings
  • ScanLogs
  • UrlDataset
  • TrainJobs
  • ModelRegistry

Purpose:

  • Persist all user data
  • Maintain complete scan logs
  • Manage training datasets
  • Track model versions and metrics

All system decisions are fully auditable.

3. AI Engine (Hybrid Model)​

ChildSafeNet uses a Hybrid AI detection strategy.

Random Forest Model​

  • 1000+ handcrafted URL features
  • URL entropy
  • Symbol patterns
  • Token structure
  • Suspicious TLD signals

NLP Pipeline​

Text-based analysis using TF-IDF or similar features.

Focus areas:

  • Adult content
  • Gambling content
  • Suspicious page language

Hybrid Policy Layer​

Combines AI scores and applies category thresholds.

The engine returns:

  • Label
  • Confidence score
  • Decision (Allow / Warn / Block)

4. Browser Extension (Manifest V3)​

Responsibilities:

  • Detect tab navigation
  • Capture URL metadata
  • Send scan requests to API
  • Enforce blocking decisions
  • Display block.html when required
  • Sync with parent protection mode

The extension never makes the final decision locally. All enforcement depends on the API response.

5. Data Flow — Scan Process​

1

Extension or Web sends scan request:

  • URL
  • Title
  • Text
  • Source

Endpoint: /api/scan

2

API processing:

  • Apply whitelist/blacklist
  • Invoke AI prediction
  • Apply parent protection policies
3

API stores results:

  • ScanLogs
  • UrlDataset entries

Decision returned to extension.

6. Option Periodic — Model Training​

Step 1 — Continuous Collection​

The API continuously stores scanned URLs inside UrlDataset.

Initial state: Pending

Step 2 — Admin Review​

Admins review dataset samples.

Options:

  • Approve
  • Reject

Only approved samples are used for training.

Step 3 — Scheduled Training​

On scheduled intervals:

  • Merge baseline dataset + approved samples
  • Train Hybrid AI model
  • Evaluate metrics:
    • Accuracy
    • Precision
    • Recall
    • F1 Score

Step 4 — Safe Deployment​

New model versions are stored in ModelRegistry.

Deployment rules:

  • Previous version kept as fallback
  • Quality gates must pass
  • Rollback possible if regression detected

7. Design Principles​

Secure
Role-based access control
Auditable
Complete scan logging
Stable
Safe model retraining
Fast
Low-latency inference

Summary​

ChildSafeNet architecture ensures:

  • Real-time threat detection
  • AI-powered classification
  • Full auditability of decisions
  • Safe model lifecycle
  • Controlled retraining strategy