This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend and Family

What I Built
I built OmniGuard AI — a free, open-source tool that detects scams, phishing messages, and fake UPI payment requests before your family falls for them.

I made it for my parents and grandparents. They keep getting fake WhatsApp messages like "Your bank account is blocked, click here immediately." They never know if it is real or a scam. Now they just paste the message and OmniGuard AI tells them instantly — is it safe or dangerous, why it was flagged, and exactly what to do next.

It works in 10 languages including Hindi and Marathi so my family can use it in their own language. 🇮🇳

Demo
🔗 https://kavach-web-2mkc.onrender.com

Code

Anurag-tech22 / DEV1

OmniGuard AI

OmniGuard AI is an open-source fraud detection and cyber defense platform. It is designed to identify and neutralize social engineering attacks, phishing, and financial fraud across digital channels.

Features • Architecture • Installation • API • Contributing • License

Features

• Threat Core: Pattern recognition for detecting digital threats, fake KYC, lottery bait, APK droppers, and courier imposter schemes.
• Threat Radar: Live scam feed, homoglyph link inspection, and QR/UPI debit analyzer.
• Audio Guard: Audio waveform visualizer and speech analyzer for voice cloning detection.
• Android APK Sandbox: Manifest privilege analyzer, Accessibility hijacking detector, and trojan profiler.
• Dossier Export: Formats incident data into structured evidence dossiers for legal submission.
• Multi-language Support: Localization in English, Hindi, Marathi, Spanish, French, German, Chinese, Japanese, Arabic, and Portuguese.
• Interactive Drills: Scenario simulator for training users to identify phishing and authority impersonation.
• Breach Intel: Checks compromised credentials, leaked…

View on GitHub

How I Built It
Backend: FastAPI (Python) with a custom heuristic scam detection rule engine
Frontend: React 19 + Vite — message scanner, live threat radar, practice mode, and a downloadable threat report you can file with the police
Deployed on Render using Docker multi-stage build — Node builds the React app, Python serves the backend, one single container
For AI, I built a custom open-source AI agent harness called OmniAIEngine. It automatically routes threat analysis to the best available model:

Local open-source inference — runs Qwen 2.5 1.5B via llama.cpp fully self-hosted, zero API cost
Falls back to Gemini or OpenAI if API keys are provided
If nothing is available, uses the built-in heuristic rule engine
This means the app works completely offline with open-weight models and needs no paid API at all.

Why Does Open Innovation Matter?
My family cannot afford a paid security product. With open source, I built something completely free that anyone in India can use. No subscriptions, no black box — you can see exactly why something was flagged, which builds real trust with non-tech users.

Open-weight models like Qwen running via llama.cpp made it possible to run real AI inference without sending anyone's private messages to a third-party cloud. Privacy-first and free — that is what open innovation made possible here.

Prize Categories
🏆 Render — Best Use of Render