Insightify

Stay Away, Stay Safe

Comment

Inspiration

Scams used to be easy to spot—bad grammar, fake logos, or clumsy emails. But today, AI has changed the game.

We were inspired to build Insightify after seeing how dangerous scams have become. People are now losing their life savings to AI voice clones that sound exactly like their relatives, deepfake videos of celebrities endorsing fake investments, and hyper-realistic phishing sites.

We realized that traditional antivirus apps are outdated. They only check for bad links. They cannot "listen" to a phone call or "watch" a video. We wanted to build a guardian that uses the same powerful AI technology to protect people, rather than deceive them.

What it does

Insightify is the world's first Multimodal Scam Detector. It works like a 24/7 digital bodyguard on your phone.

  • Multimodal Analysis: You can upload Screenshots, Audio recordings, Videos, or paste Text.
  • AI Reasoning: It doesn't just look for keywords. It "understands" the context. For example, if an audio clip sounds urgent and asks for money, Insightify flags it as a "High Risk" scam, even if no malicious link is sent.
  • Real-Time Protection: It scans incoming SMS and notifications instantly.
  • Educational Insights: It doesn't just say "Block this." It explains why (e.g., "This message uses a 'Pig Butchering' scam pattern common in crypto fraud").

How we built it

We built Insightify using React Native for a smooth mobile experience and a Node.js/Express backend hosted on Railway.

The core brain of our app is Google Gemini.

  1. Gemini 3 Flash (Preview): We use Google's latest model for complex reasoning. It analyzes the tone of voice in audio files and detects subtle social engineering cues in text.
  2. Gemini 1.5 Flash: We implemented a "Smart Fallback" system. If the experimental Gemini 3 model is overloaded (Error 503), our backend automatically switches to the stable Gemini 1.5 Flash model. This ensures our users are never left unprotected.
  3. Lottie Animations: We used vector animations to create an engaging and friendly user interface.

Challenges we ran into

  • Handling Large Files: Uploading video and audio files from a mobile device to a server is tricky. We had to implement a robust file-handling system that converts media to Base64 chunks without crashing the app.
  • Model Overload: During testing, the Gemini 3 Preview model was sometimes busy due to high traffic. This broke our app initially.
  • Solution: We wrote a custom "Auto-Fallback" algorithm in our backend. It tries the newest model first, but instantly switches to a backup model if it detects an error. This made our app "crash-proof."

Accomplishments that we're proud of

  • True Multimodality: We successfully got the AI to "watch" a video and "listen" to an audio file and accurately detect a scam within seconds.
  • Smart Fallback System: Building a backend that intelligently switches between AI models to guarantee uptime.
  • The UI/UX: Creating a clean, professional interface that makes complex AI security tools feel simple and friendly for non-technical users (like our grandparents).

What we learned

  • We learned that Context is King. A message saying "Send money" is fine between friends, but dangerous if it comes from an unknown number. Gemini's ability to understand this context is what makes it superior to traditional code-based filters.
  • We mastered Prompt Engineering. Telling the AI to act as a "Forensic Cybersecurity Analyst" gave us much better results than just asking "Is this a scam?".

What's next for Insightify

  • Live Call Protection: We plan to implement real-time call screening that listens to phone calls live and alerts you if a scammer is speaking.
  • Browser Extension: Bringing Insightify to the desktop to protect users while they browse the web.
  • Community Reporting: Allowing users to submit new scams to a shared database to warn others in their area.

Built With

Share this project:

Updates