NetraEdge v1.0.0 — NHAI Innovation Hackathon 7.0

NetraEdge v1.0.0 — NHAI Innovation Hackathon 7.0

Offline facial recognition and 10-layer liveness detection for zero-network environments. Solo developer submission.

Highlights

• 10-layer algorithmic liveness fusion — all layers are signal-processing (LBP, HSV, FFT, Laplacian, optical flow, gradient histograms, sensor fusion, blendshapes, POS rPPG). The CNN (liveness_detector.tflite, 0.52 MB) is shipped as a bypassed backup only.
• 27–33 ms per-frame inference on a mid-range Motorola G-series (Adreno 610, 4 GB RAM)
• End-to-end verify < 1 s (3-second grace period + 2 active challenges + 128-d cosine match)
• MobileFaceNet 128-d at 99.48% LFW accuracy (Apache-2.0, paper benchmark)
• AES-256-GCM at rest + TLS 1.3 in transit + differential privacy (Laplace, ε=1.0) on synced embeddings
• Datalake 3.0 integration with HTTPS sync, exponential backoff, and signed audit log
• 121 / 121 unit tests passing, 0 lint errors, 0 typecheck errors in CI

APKs (debug-signed, installable)

File	Size	Architecture	Use this if…
�pp-arm64-v8a-debug.apk	41.83 MB	arm64-v8a	…your device is 2018+ (Snapdragon 660+, Exynos 9+, MediaTek Helio P60+, all Apple A11+)
�pp-armeabi-v7a-debug.apk	34.15 MB	armeabi-v7a	…your device is 2014–2017 budget Android
�pp-universal-debug.apk	92.45 MB	both	…you want one file that works everywhere

SHA256 (arm64): a4d9dc1a034b4dab9b566f4287d30f40a2b2fae6aba9e5ebfbfc4ca8a53007c1 SHA256 (armv7): de33b24f0f1f4565e1c102b35e5985a5613938f63ec039d10f5e69239a15444c SHA256 (universal): 2d80508f2193dbf1435d65e0bf6cbf7e360c3c679a8571abe9bf1734c7d1c414

Install

`�ash

Pick the right ABI for your device, then:

adb install -r -t -g app-arm64-v8a-debug.apk
adb shell am start -n com.netraedge/.MainActivity
`

What's inside

• 20 Kotlin files (~4,900 LOC) implementing the on-device 10-layer pipeline
• 65 TypeScript + 14 TSX files (cross-platform React Native layer)
• 1 Swift file (iOS bridge, 386 LOC)
• 9 markdown docs (architecture, API, benchmarks, integration, model card, hackathon submission, presentation deck, acknowledgments)
• 14-slide PPTX presentation deck (372+ native shapes, 2026 dark theme)
• 3 benchmark JSON files (latency, accuracy, liveness)
• AWS SAM template (API Gateway + Lambda + DynamoDB)
• Express mock sync server + AWS Lambda production handler

Demo flow

1. Launch the app — 2026 premium UI: glassmorphism, conic gradients, particle effects
2. Tap Enroll — align face in green reticle, complete 2 random active challenges (BLINK / SMILE / HEAD_TURN), 10 layers verify liveness, 128-d embedding encrypted and stored locally
3. Tap Verify — 3-second grace period, 10 layers evaluate liveness, 2 active challenges, embedding matched at cosine ≥ 0.62, VERIFIED badge appears with golden glow
4. Try spoofing — show photo / replay video / wear mask — SPOOF badge in red with forensic detail
5. Tap Sync — HTTPS POST to Datalake 3.0 endpoint, 200 OK, local cache auto-purged, signed audit log retained
6. Long-press Sync — endpoint override dialog (for restricted networks)

Compliance

Standard	Status
NHAI 7.0 (a) offline liveness	✅ 10 layers, all algorithmic
NHAI 7.0 (b) sync & purge	✅ 6-step flow, signed audit log
DPDP Act 2023 (Sections 6, 8, 11, 17)	✅ Consent, purpose limitation, erasure, audit
Apache-2.0 only (no NC/ND)	✅ MediaPipe, MobileFaceNet, MiniFASNet
MIT license	✅ Source code

Acknowledgments

MediaPipe (Google) · MobileFaceNet authors (Chen et al., 2018) · MiniFASNet authors (Yu et al., 2020) · Wang et al. (2016) for the POS rPPG algorithm · Erdoğmuş & Marcel (2014) for the 3D-mask attack taxonomy.

---

NHAI Innovation Hackathon 7.0 — Datalake 3.0 · Submission deadline 05.06.2026 23:59 IST