PhishGuard AI

🔹 “Detect scams before they catch you.” 🔹 “Stay aware. Stay secure.” 🔹 “Smart defense against digital deception.” 🔹 “See the scam before you click.” 🔹 “Protecting your trust in a digital world.”

Comment

Inspiration

With the rise of UPI payments, digital banking, and online communication, millions of people are becoming victims of smart social engineering scams, especially through WhatsApp, SMS, fraudulent calls, and fake QR codes. Traditional antivirus software can’t detect these threats because they’re not viruses—they’re psychological manipulations powered by technology. We wanted to build a system that protects users not from code-based threats, but from human deception using digital tools.

What it does

PhishGuard AI is an AI-powered fraud detection shield that:

Analyzes messages, links, PDFs, invoices, QR codes, and voice calls for scams.

Detects deepfake scam calls that impersonate officials, banks, or family.

Predicts user vulnerability using behavior-based risk scoring.

Provides instant safety actions and explanations in simple language.

Generates a cyber evidence report to help victims file legal complaints easily.

Simply put, PhishGuard AI warns users before they become victims. We built PhishGuard AI using a combination of AI models, cybersecurity logic, and no-code orchestration through Base44.

Base44 helped us create the workflow logic for:

Document/image/voice data routing

Triggering cybersecurity alerts

Automating report generation steps

Connecting phishing analysis modules without manually coding every pipeline

FastAPI + Python powers our backend inference engine

React/Tailwind handles the dashboard and interactive alerts

NLP models (BERT/RoBERTa) detect phishing and scam text patterns

Tesseract OCR + YOLO extract and analyze fake invoices, IDs, and QR scams

Whisper + MFCC + CNN identify scam calls and deepfake voice manipulation

Behavioral analytics predict user vulnerability based on hesitation, urgency response, and risky clicks

Together, Base44 automation + custom AI models allowed us to build a multi-modal fraud protection system rapidly while keeping it scalable and practical for real-world use.

Challenges we ran into

Challenges we ran into

Designing a model that detects social engineering tactics, not just malicious links.

Combining multi-modal analysis (Text + Image + Audio + Behavior).

Detecting deepfake voice scams on low-quality call recordings.

Creating warnings that are simple, accurate, and non-panic inducing for beginners.

Balancing security with real-time speed on limited devices.

Accomplishments that we're proud of

Built a multi-format fraud detector that works across text, voice, QR, and PDFs.

Designed a behavioral risk model rarely used in consumer security tools.

Implemented cyber evidence reporting, making legal complaints easier.

Developed a minimal, user-friendly interface that educates without overwhelming users.

What we learned

Cybersecurity is no longer just technical—psychology + AI = real protection.

Deepfake detection in voice needs to consider tone patterns, pauses, and background noise, not just audio frequency.

Users respond better to calm suggestions vs. fear-based warnings.

Real-world security tools must be explainable and trustworthy, not just accurate.

What's next for PhishGuard AI

Launch a browser extension for real-time URL and QR scanning.

Integrate with UPI apps and banks to flag fraudulent payment links instantly.

Offer enterprise training dashboards for employees and students.

Build a WhatsApp fraud alert bot to scan forwarded messages before they spread.

Add regional language support for scam call detection.

Built With

  • firebase
  • firebase-(auth-&-realtime-notifications)-|-|-**cloud-deployment**-|-vercel-(frontend)
  • gpt-api-|-|-**voice-&-deepfake-analysis**-|-whisper-api
  • huggingface-(ai-models)-|-|-**cyber-evidence-report**-|-reportlab-(pdf-generator)
  • javascript
  • mfcc-+-cnn-models-|-|-**ocr-+-vision-for-pdf/qr**-|-tesseract-ocr
  • node.js-(for-auxiliary-services)-|-|-**mobile/web-automation-(no-code)**-|-**base44**-(workflow-automation-&-pipeline-orchestration)-|-|-**text-phishing-ai-(nlp)**-|-bert
  • render/cloud-run-(backend)
  • roberta
  • storage
  • tailwind-css
  • typescript-|-|-**frontend**-|-react.js
  • vite-|-|-**backend**-|-fastapi-(python)
  • yolov8-|-|-**database**-|-mongodb
  • |-category-|-technology-|-|-|-|-|-**programming-languages**-|-python
Share this project:

Updates