DesiCheck - The AI Scam Shield

A multimodal AI bodyguard fighting digital scams and misinformation for the common man in India.

Comment

💡 Inspiration

In India, "WhatsApp University" and digital scams are a crisis. Every day, millions of seniors, students, and regular citizens fall victim to phishing links, fake job offers, and predatory loan apps.

My grandfather can't analyze a URL header. My friends often panic when they see "Urgent: Your account will be blocked" messages.

I realized that while corporations have massive Security Operation Centers (SOCs), the common man has no one. I built DesiCheck to fill this gap. It is an Agentic AI Bodyguard that doesn't just "detect" problems but explains them in simple language and even handles the situation for the user.

🛡️ What It Does

DesiCheck is a Progressive Web App (PWA) acting as a personal digital security expert offering three core defenses:

  1. AI Screenshot Analysis (The Scam Triage): Users can upload screenshots of suspicious messages (WhatsApp, SMS, Email). The app uses Gemini 1.5 Flash (Multimodal) to analyze visual cues (blurry logos, urgent fonts) and text content to detect social engineering tactics. It returns a verdict (Safe/Scam) with a "Risk Score."
  2. Deep Link Inspection: Users paste a suspicious link, and the system acts as a forensic tool. It checks for typosquatting (e.g., hdfcc-bank.com), domain age, and known malicious patterns.
  3. Viral Counter-Measure (The Troll Agent): If a scam is detected, the AI generates a safe, sarcastic, and context-aware reply (in Hinglish) that the user can send back to the scammer to waste their time without revealing personal info.

⚙️ How We Built It

The project is built on a modern, serverless architecture to ensure zero maintenance and high scalability:

  • Frontend: Vanilla HTML/CSS/JS. We chose this for its lightweight footprint, ensuring fast loading on Indian mobile networks.
  • AI Engine: Google Gemini 1.5 Flash API. We used advanced System Prompting to instruct the model to act as a "Ruthless Security Analyst," prioritizing safety and explaining technical risks in layman's terms.
  • Development Environment: Built entirely inside Project IDX, Google's AI-assisted IDE, which accelerated the coding process.
  • Deployment: Containerized with Docker and deployed to Google Cloud Run. This ensures the app scales automatically—whether 10 people use it or 10,000.

🧠 Challenges We Ran Into

  • Prompt Engineering for Context: Getting the AI to understand "Indian contexts" (like specific UPI scams or "free recharge" schemes) required fine-tuning the system prompts.
  • Reducing Latency: Since we are using an AI model, the response time was initially slow. We optimized the image compression before sending it to the API to speed up the "Verdict" phase.
  • Cloud Run Configuration: Learning to containerize a vanilla JS app and deploying it stateless on Cloud Run was a steep learning curve, specifically regarding Dockerfiles and Port configurations.

🏅 Accomplishments That We're Proud Of

  • Successfully integrated Multimodal AI (Text + Vision) in a single workflow.
  • Achieved a serverless deployment on Google Cloud Run that creates a professional https:// endpoint.
  • Created a tool that is genuinely useful for my own family members to vet suspicious messages.

🚀 What's Next for DesiCheck

  • Voice Mode: Adding Audio Input/Output for users who cannot read/write well.
  • Regional Language Support: Expanding the output to support Bengali, Tamil, and Telugu.
  • Community Report: Creating a shared database where verified scams are stored to alert other users instantly.

Built With

Share this project:

Updates