AutoPilot AI: Agentic Phishing Defense

• 

  Microsoft 365 Excel Logged

• 

  n8n Workflow

🛡️ Inspiration

In a world where phishing attacks are becoming more sophisticated, simple keyword filters are no longer enough. Living in Nigeria, I have seen how cyber threats can impact students and small businesses. I was inspired to build AutoPilot AI—a real-time security agent that doesn't just block emails, but actually reasons through them like a human security analyst.

🚀 What it does

AutoPilot AI monitors a user's Gmail in real-time. When a new email arrives, it is processed by Amazon Nova Lite, which performs a deep forensic analysis of the sender's intent, urgency, and suspicious links. The system then:

• Tiers the Risk: Categorizes the threat as High, Medium, or Low.
• Logs the Logic: Writes the AI's full reasoning into a Microsoft 365 Excel ledger for review.
• Automates Triage: Saves security teams hours of manual work by highlighting only the most critical threats.

🛠️ How I built it

I built the "brain" of the agent using Amazon Nova Lite via AWS Bedrock. For the "nervous system," I used an n8n automation hub to connect multiple APIs. This allowed me to create a seamless agentic workflow that bridges the gap between Gmail triggers and real-time AI analysis. I focused on making the architecture reliable to ensure 24/7 phishing monitoring.

🧠 Challenges I faced

• Reducing Noise: Early versions flagged too many "False Positives" (like short personal emails). I solved this by refining the prompt logic to use a Tiered Risk System, which reduced notification fatigue.
• Infrastructure Reliability: Setting up a consistent environment for the agent to run 24/7 was a challenge. I overcame this by deploying a cloud-hosted instance of my automation workflow to ensure it remains active even when my local machine is offline.
• Security: Ensuring that the workflow remained secure while handling sensitive email metadata required careful OAuth2 credential management and the use of environment variables to protect API keys.

🎓 What I learned

This project taught me the power of Agentic AI. Instead of a linear script, I learned how to build a system that can make autonomous decisions based on the content it sees. It also deepened my skills in API orchestration and the importance of choosing the right model—like Amazon Nova Lite—for high-speed, cost-effective reasoning. #AmazonNova

Built With

• bedrock
• google-gmail-oauth
• javascript
• microsoft
• n8n
• nova