-
-
Architecture
Inspiration
I built Pay Guardian for everyday people. I believe protecting yourself from social engineering scams should require zero technical knowledge or experience, because everyone deserves to be safe online.
I wanted to create a simple tool that acts as a last line of defence before someone follows harmful instructions hidden in a PayPal invoice.
What it does
Pay Guardian is a Chrome extension that uses AI to identify potential social engineering tactics in PayPal invoices.
It currently detects urgency and pressure, and requests for sensitive information. It explains suspicious instructions in plain language and provides practical safety guidance.
How we built it
I built Pay Guardian using a Chrome extension, Next.js, TypeScript, Google Gemini and the Vercel AI SDK.
The extension extracts invoice notes and sends them to a hosted backend for AI analysis.
I also used PayPal's Invoicing API to programmatically generate sandbox invoices covering different test scenarios.
Challenges we ran into
The biggest challenges were reliably extracting invoice content and distinguishing suspicious instructions from legitimate messages to avoid unnecessary warnings.
Creating realistic test scenarios was also time-consuming, which led me to automate invoice generation using PayPal's API.
Accomplishments that we're proud of
I'm proud of building a working solution that requires no technical expertise to use.
Guardian can identify multiple warning signs in a single invoice, provide clear explanations, and offer safety guidance when no obvious threat is detected.
What we learned
I gained experience developing Chrome extensions, integrating AI with structured outputs, and working with PayPal's Developer API.
I also learned how important context is when detecting social engineering, and why security warnings must be both accurate and easy to understand.
What's next for Pay Guardian
I plan to expand detection to include impersonation, suspicious links and misleading payment instructions.
I'll also extend analysis beyond invoice notes, improve detection accuracy through larger test datasets, and prepare Guardian for wider public use.
My goal remains the same: make online safety accessible to everyone, regardless of technical experience.
