I wanted to solve a specific problem: AI Agents are becoming autonomous, but we have no visibility into their security "conversations" in real-time.
To fix this, I built Vanguard_SOC, a live security operations dashboard that intercepts prompts, flags malicious intent (SQL Injection, PII harvesting, Social Engineering), and generates immutable audit reports.
🛡️ Core Features:
Live Forensics: Real-time data streaming via Firebase Firestore ensures every prompt is audited the millisecond it happens.
Threat Detection Engine: Automated pattern matching for credit card details, SSNs, and malicious SQL commands.
Compliance Reporting: One-click PDF export for security auditors using jspdf-autotable.
Zero-Trust UI: High-contrast dashboard designed for ...
I wanted to solve a specific problem: AI Agents are becoming autonomous, but we have no visibility into their security "conversations" in real-time.
To fix this, I built Vanguard_SOC, a live security operations dashboard that intercepts prompts, flags malicious intent (SQL Injection, PII harvesting, Social Engineering), and generates immutable audit reports.
🛡️ Core Features:
Live Forensics: Real-time data streaming via Firebase Firestore ensures every prompt is audited the millisecond it happens.
Threat Detection Engine: Automated pattern matching for credit card details, SSNs, and malicious SQL commands.
Compliance Reporting: One-click PDF export for security auditors using jspdf-autotable.
Zero-Trust UI: High-contrast dashboard designed for SOC (Security Operations Center) monitoring.
🛠️ The Tech Stack:
Frontend: Next.js 16 (App Router) & Tailwind CSS
Backend: Firebase (Firestore & Cloud Functions)
Deployment: Netlify with a CI/CD pipeline from GitHub
Reporting: jsPDF
🚀 Lessons Learned:
Building this taught me that the biggest threat to Agentic AI isn't just the logic, but the social engineering attempts used to trick agents into revealing temporary codes or resetting credentials.
Check out the live dashboard here: https://vanguard-soc.netlify.app/
I’d love to hear from the security community—what specific "Prompt Injection" patterns should I add next to the detection engine?