VIGILANT AI — Intelligent Threat Detection & Cybersecurity

Inspiration

Scams are becoming increasingly sophisticated. Fraudulent bank messages, fake delivery notifications, job offer scams, business email compromise, and attacks targeting AI agents can manipulate people into making unsafe decisions. Traditional warning signs are not always obvious, especially when attackers use urgency, authority, fear, or convincing impersonation.

I wanted to build a tool that helps people recognize these threats and understand why a message may be dangerous, rather than simply labeling it as safe or suspicious.

What It Does

VIGILANT AI is an AI-assisted cybersecurity and threat-analysis platform designed to help users identify suspicious messages and understand potential security risks.

The platform analyzes submitted text and generates a threat report containing:

• Risk scoring: A score out of 100 to communicate the estimated level of risk.
• Threat classification: Categorizes risk into levels such as moderate, high, or critical.
• Evidence-based explanations: Highlights suspicious language and signals contributing to the assessment.
• Manipulation detection: Helps identify pressure tactics and patterns commonly associated with social engineering.
• Safety recommendations: Provides guidance to help users respond more cautiously to suspicious messages.
• Interactive demonstrations: Includes examples of bank SMS scams, delivery phishing, executive business email compromise, job offer scams, and AI agent injection attempts.

My goal is to make cybersecurity analysis more understandable and actionable, even for people without a technical security background.

How I Built It

I developed VIGILANT AI as a web application using TypeScript, React, and a Node.js/Express backend. Vite supports the development workflow, while Vitest is used for automated testing.

I integrated Google's Gemini API for AI-assisted analysis and implemented a deterministic local analysis engine as a fallback when the external AI service is unavailable. This allows the application to continue generating threat reports without depending entirely on an external model.

During testing, I verified that the interactive threat demonstrations generate reports and produce different risk scores for different scenarios.

Challenges I Faced

One of my main challenges was handling external AI service limitations. When the Gemini free-tier quota was exhausted, live model requests became unavailable. Instead of leaving the application unusable, I worked on fallback handling and a visible status indicator to communicate when local analysis was active.

Another challenge was calibrating threat scores so different attack scenarios could be distinguished and the results remained understandable. I also tested the application to ensure the dashboard could generate reports consistently.

These challenges taught me the importance of resilience, transparency about system limitations, and testing beyond the visual interface.

What I Learned

Building VIGILANT AI strengthened my understanding of full-stack development, threat-analysis workflows, API integration, error handling, automated testing, and designing security tools that explain their results.

I learned that an effective cybersecurity tool should not only flag suspicious activity but also help people understand warning signs and make safer decisions.

Future Improvements

I plan to improve the platform by adding broader message-format support, integrating additional threat intelligence, evaluating performance against labeled datasets, and measuring detection accuracy and false-positive rates.

VIGILANT AI aims to make digital threat awareness more accessible by turning suspicious messages into understandable risk reports and practical safety guidance.

Built With

• api
• artificial-intelligence
• cybersecurity
• express.js
• fraud-detection
• generative-ai
• google-gemini
• node.js
• phishing-detection
• react
• social-engineering
• threat-detection
• typescript
• vite
• vitest