DataLake Guard v2.0

🎥 Demo Video: Watch the Demo on YouTube

A production-grade offline facial recognition and liveness detection system for NHAI field personnel authentication. Works 100% offline in zero-network zones and syncs to AWS when connectivity is restored.

• Real face recognition using MobileFaceNet TFLite (INT8 quantized)
• Platform-specific detection: MLKit (Android) + Apple Vision (iOS)
• Active liveness detection — blink + head turn (anti-spoofing)
• 100% offline operation — works in airplane mode
• SQLite encrypted local storage with WAL mode
• AWS REST sync + purge when network restored
• Hindi + English bilingual UI
• Government-grade NHAI design system
• Personnel directory with search
• Calendar attendance view

graph TD
    subgraph MC["Mobile Client (Offline Edge)"]
        A["Camera Frame"] --> B["Native Detection Bridge"]
        
        subgraph PD["Platform Detection"]
            B --> B1["Android: Google MLKit"]
            B --> B2["iOS: Apple Vision Framework"]
        end
        
        B1 --> C["Active Liveness Validation"]
        B2 --> C
        
        subgraph LC["Liveness Checks (Anti-Spoofing Challenge)"]
            C --> C1["Eye Aspect Ratio / EAR Blink Check"]
            C --> C2["Euler Yaw Angle Head Turn Check"]
        end
        
        C1 --> D["Embedding Extraction"]
        C2 --> D
        
        D --> D1["MobileFaceNet TFLite Inference"]
        D1 --> E["Matching Engine"]
        E --> E1["Cosine Similarity Score >= 0.72"]
        E1 --> F{"Auth Decision"}
        F -->|GRANTED| G["Write Log to SQLite WAL"]
        F -->|DENIED| H["Display Access Denied"]
    end

    subgraph SS["SQLite Storage (AES-256)"]
        G --> G1[("Encrypted Database")]
    end

    subgraph SQ["Synchronization Queue"]
        G1 --> I["NetInfo Network Listener"]
        I -->|Online Detected| J["Outbox Queue Processor"]
    end

    subgraph CB["Cloud Backend (AWS REST)"]
        J -->|Secure HTTPS Batch| K["AWS API Gateway"]
        K --> L["AWS Lambda"]
        L --> M[("Central Data Lake")]
        M -->|Sync Acknowledged 200 OK| N["Purge Local Face Embeddings"]
    end

• Face detection: < 100ms
• Liveness check: real-time (500ms snapshot loop)
• Embedding extraction: < 200ms
• Total auth pipeline: < 1 second ✅
• Model bundle size: ~1.2 MB ✅
• Minimum device: 3GB RAM, Android 8+ / iOS 13+

• Blink detection via eye open probability (Android)
• Head turn detection via Euler yaw angle (both platforms)
• Photo spoof rejected (no natural eye movement)
• Video spoof rejected (challenges are randomized)

• All data stored in SQLite (WAL mode for performance)
• Face embeddings stored as Base64 compressed strings
• Attendance logged locally with sync_status flag
• Background outbox queue uploads when WiFi restored
• Raw embeddings purged after confirmed sync

npm install npx react-native run-android

npm install cd ios && pod install && cd .. npx react-native run-ios

curl -L "https://github.com/MCarlomagno/FaceRecognitionAuth/raw/master/assets/mobilefacenet.tflite" -o assets/models/mobilefacenet.tflite

src/ ai/ - TFLite model wrappers db/ - SQLite repositories sync/ - AWS sync service screens/ - All UI screens navigation/ - React Navigation setup utils/ - Helpers and constants native/ - iOS native modules assets/models/ - TFLite model files

https://github.com/Nightkilller/DataLakeGuardV2

Solo participant — NHAI Hackathon 7.0 Submission: June 5, 2026