Eiko AI Cyber Guard

Landing Screen where User enters information about the suspicious message
Result Page Showing Feedback From Both OpenAI Chagpt and Google Gemini based on data sent
Result Page Showing Feedback For a Positive Scenario
Audio Phone Call Realtime Fraud Detection Module Transcribing and Detecting Fraudulent Phone Calls

Inspiration

In a world where online interactions are frequent, fraudsters often use social engineering tactics to manipulate users into sharing sensitive information—a growing concern reflected in global statistics, with cybercrime losses reaching an estimated $8 trillion in 2023 alone and phishing attacks accounting for nearly 30% of all breaches worldwide.

I set out to create a solution that provides immediate, reliable protection by harnessing the power of dual AI systems to help users identify and handle fraudulent messages. Inspired by the principle that "two heads are better than one," I integrated both Google Gemini and OpenAI's ChatGPT, delivering feedback from both AIs for enhanced user safety.

Additionally, our real-time phone call fraud module helps users tackle phone call fraud as it happens. While users are on calls, it leverages Google Speech-to-Text and Google Gemini to transcribe and detect potential fraud in real-time.

What it does

Eiko AI Cyber Guard contains two modules (Fraudulent Text Detection and Fraudulent Phone Call Realtime Detection).

Fraudulent Text Detection Module

it allows users to detect fraudulent messages across any communication platform. By leveraging the capabilities of Google Gemini-1.5-flash AI Model and OpenAI ChatGPT, the system provides side-by-side feedback to users, highlighting suspicious patterns and offering personalized advice on how to handle the potential threat. This dual AI approach ensures higher accuracy in identifying fraud and protecting users from scams and social engineering hacks. I was able to test out our approach in Google AI Studio using Gemini-1.5-flash.

Real-Time Fraudulent Phone Call Detection Module

This module helps users detect fraudulent calls where scammers impersonate legitimate organizations, such as banks and fintech companies, in an attempt to steal users' credentials.

It listens to phone calls and performs real-time fraud detection as the conversation progresses, utilizing the Google Speech-to-Text API and the Google Gemini-1.5-Flash AI Model API. Users receive immediate advice and fraud alerts, enhancing their protection during active calls.

How We Built It

Core Framework and Deployment

Framework: Python-JavaScript Flask app for backend service.
Deployment: Google Cloud Compute Engine VM instance is where our app was deployed.
CI/CD: GitHub for continuous integration, using Google Cloud CLI (gcloud CLI) for environment setup and deployment automation. -Prototyping: Google AI Studio Gemini 1.5-Flash Model.

Backend Integrations

AI Processing: OpenAI and Google Gemini-1.5-Flash AI Model APIs to analyze user-submitted messages
Speech-to-Text: Google Speech-to-Text API for real-time transcription of phone calls, enabling immediate fraud detection through Google Gemini-1.5-Flash AI
Real-Time Communication: flask_socketio for efficient real-time communication between frontend and backend

Frontend Interface

Real-Time Sockets: socket.io.min.js for real-time communication with the backend.
Data Collection: Captures user interactions with potential fraudsters and sends data to the backend for analysis
AI Response Display: Displays feedback from both AI models side-by-side for user insights using HTML, CSS, Javascript.

Eiko ai system design

Challenges we ran into

One of the biggest challenges was optimizing the integration between two AI models—Google Gemini and OpenAI’s ChatGPT. Ensuring that the feedback was cohesive and timely was crucial to creating a smooth user experience. Additionally, designing a system that handles real-time fraud detection from audio phone calls gave me a lot of technical challenges but I was able to resolve them and learn a lot from the development and deployment process.

Accomplishments that we're proud of

I am proud of the successful dual integration of Google Gemini and OpenAI ChatGPT, which has allowed us to provide users with two separate but complementary perspectives on message safety. The feedback users receive is both personalized and actionable, offering them real-time advice to safeguard against fraud during a phone in real-time is something I am very excited about discovering. Our two-AI approach stands out as a unique and powerful aspect of our project.

What we learned

This project deepened our understanding of how to integrate multiple AI models into a single product. We learned valuable lessons about API optimization, asynchronous handling of multiple API calls, and designing an intuitive user interface that could handle real-time data in a user-friendly way.

What's next for Eiko AI Cyber Guard

Looking ahead, we plan to expand Eiko AI Cyber Guard by incorporating additional AI models to further enhance the accuracy and reliability of our fraud detection. We are also considering partnerships with messaging platforms and email services to provide seamless, real-time fraud detection directly within those systems. Lastly, we aim to refine the user experience, making the system faster, more responsive, and even easier to use.