GhostNet AI advances from an agile hackathon prototype to an enterprise-grade cyber defense intelligence platform.
- Feature 1 — Explainable Risk Scoring: 10 standardized reason codes (
URGENCY_SCARE_TACTICS,REQUEST_OTP_PASSWORD, etc.) with code-defense parser. - Feature 2 — Hardened Deterministic Offline Heuristics: Offline regex and NLP scoring engine with
withTimeout(promise, 9000)protection. - Feature 3 — Public Threat Feed Synchronization: Automated daily cron ingestion of OpenPhish and URLhaus with SHA-256 domain hashing.
- Feature 4 — Community Threat Intelligence Telemetry: Supabase PostgreSQL
threat_indicatorsdatabase with atomic PostgREST upsert (on_conflict=indicator_hash). - Feature 5 — Live QR Code Camera Scanner: Real-time WebRTC camera viewfinder and canvas decoding via
jsQRwith fallback image uploading. - Feature 6 — Voice & Deepfake Call Detector: In-memory Fast Fourier Transform (
fft.js) spectral flatness (Wiener entropy) and Groq Whisper-large-v3 transcription with HTTP 413 (>6MB) guards. - Feature 7 — Real-Time Chromium Browser Extension (MV3): Manifest V3 extension featuring background tab monitoring, automated hash checks, and warning overlays.
- Feature 8 — Pre-Click Interceptor & Sandbox: Low-latency
/api/blocklist-litestreaming API and zero-execution educational link sandbox. - Threat Reconstruction™: Interactive 5-stage cyber kill-chain mapper (
Ingress➔Social Engineering➔Phishing Gateway➔Credential Harvesting➔Loss). - Attacker Intent Engine: Plain-English adversary motivation translation.
- Family & Senior Safety Mode: Enlarged typography, high-contrast palette, and simplified non-technical advice.
- Instant Theme Engine: Persistent Light and Dark mode with ThreeUI ambient living particle fields.
- Cross-Platform Mobile: Native Android package configured and tested via Capacitor 8.5.
- Guest / Judge Demo Mode: 1-click
⚡ Explore as Guest / Judge Demo Modelogin gate bypass. - Automated Quality Gates: 40/40 tests across 15 suites in
tests/scanner.test.jspassing in <400ms.
- Android Native Share-to-Scan Target: System-level Android intent allowing users to share suspicious messages directly from WhatsApp, SMS, or Telegram without copy-pasting.
- Multilingual Regional Indian & Global Languages: Fine-tuned local NLP heuristic models for Hindi, Tamil, Telugu, Bengali, Spanish, and French.
- Automated Bank & Law Enforcement Helpline Registry: Dynamic directory connecting users directly to verified corporate hotlines (State Bank of India, HDFC, ICICI, PayPal, FedEx) and national reporting authorities.
- Firefox & Safari Extension Ports: Porting the MV3 Browser Shield to Firefox WebExtensions and Safari App Extensions.
- On-Device Small Language Model (SLM) via WebGPU: Quantized on-device 1B-parameter threat classification running entirely in client hardware using WebGPU and MediaPipe for 100% private, zero-cloud scans.
- Decentralized Threat Blockchain Oracle: Cryptographically signed threat indicators syndicated across participating enterprise security vendors with zero centralized control.
- Enterprise SOC Multi-Tenant Dashboard: Organization-wide phishing simulation, workforce threat awareness analytics, and SIEM/SOAR API webhooks.