AWS Builder Center
Building FakeBuster: A Multimodal AI Shield Against Digital Fraud using AWS Bedrock

Building FakeBuster: A Multimodal AI Shield Against Digital Fraud using AWS Bedrock

How Team Gawds built a 3-in-1 AI scanner using Amazon Bedrock, S3, and Transcribe to protect users from phishing, forged invoices, and vishing calls during the AWS First Commit Hackathon.

Digital deception in India is evolving faster than ever. Scammers are no longer just sending suspicious SMS links—they are forging UPI payment screenshots, sending fake bank invoices, and using AI voice cloning to impersonate authorities in highly orchestrated "vishing" (voice phishing) attacks.
During the AWS First Commit Hackathon, our team (Team Gawds) wanted to build a centralized, enterprise-grade shield to combat this. We built FakeBuster, a multimodal AI web application that detects digital fraud in under 5 seconds.

🚀 What We Built

FakeBuster is a 3-in-1 scanner. Users can:
  1. Paste Text: Analyze suspicious SMS, WhatsApp forwards, or URLs.
  1. Upload Images: Scan payment receipts or screenshots for digital manipulation.
  1. Upload Audio: Submit voice notes or recorded calls to detect vishing and AI voice cloning.
For every scan, the app returns a Threat Score (0-100), an AI-generated analysis report, and actionable red flags.

🛠️ The AWS Architecture

To make this possible, we heavily leveraged AWS Cloud Services:
  • Amazon Bedrock (Claude 3 Haiku): This was our core AI reasoning engine. We engineered specific prompts to look for psychological manipulation (urgency, authority impersonation) and inconsistencies. Bedrock's unified API made it incredibly easy to integrate multimodal analysis for both our text and image scanners.
  • Amazon S3: We used S3 as our secure media storage layer. When users upload audio recordings of suspicious calls, they are temporarily bucketed here.
  • Amazon Transcribe: We integrated Transcribe into our audio pipeline. It accurately transcribes the uploaded voice calls (even handling Indian accents wonderfully), which we then pass back to Bedrock to analyze the transcript for vishing tactics.
  • Frontend: Built with Next.js (TypeScript) and styled with Tailwind CSS, deployed on Vercel. We designed a premium light theme with glassmorphism and a custom SVG threat gauge to ensure the user experience feels trustworthy and professional.

🛑 Challenges & What We Learned

Our biggest hurdle wasn't technical—it was administrative. As students in India under AISPL (Amazon Internet Services Private Limited), we ran into strict billing and credit card verification limitations that initially blocked our access to Bedrock foundation models in the  us-east-1  region. We had to architect a mock-response fallback system in our API routes to ensure our frontend UI and demo flow remained perfectly functional while resolving the account firewall.
This taught us a massive lesson in resilient system design. Even when backend cloud services are temporarily inaccessible due to permission errors or outages, the frontend must degrade gracefully and maintain a polished user experience.

🎉 Conclusion

Building FakeBuster was an incredible experience. Integrating Amazon Bedrock, S3, and Transcribe opened our eyes to how fast you can build enterprise-grade AI applications when using the right cloud infrastructure. We are excited to continue refining FakeBuster to protect vulnerable people from the next wave of AI-generated scams.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article