🛡️ Guardian Agent: Self-Healing AI Triage

This is a lightweight AI microservice built to demonstrate resilient, unattended LLM workflows.

In the real world, LLMs frequently hallucinate or fail to return strict JSON formats, which breaks downstream systems. This project solves that by implementing a Self-Healing Try-Catch Loop.

🚀 Key Features Demonstrated (Aligning with AI Web Developer Role)

• Failure Recovery: If the AI returns malformed JSON, the system intercepts the SyntaxError, dynamically rewrites the prompt to include the error log, and forces the model to auto-correct itself.
• Unattended Automation: Once deployed, it can process webhooks or user inputs infinitely without human intervention, logging its success/failure rates to a local SQLite database.
• Full-Stack Integration: Uses an Express.js server to bridge the gap between a simple front-end UI and complex back-end AI orchestration.

🛠️ The Tech Stack

• Backend: Node.js, Express
• AI: OpenAI API (gpt-4o)
• Database: SQLite3 (for auditing and state management)
• Frontend: Vanilla HTML/CSS/JS

💡 Why I Built This

While my core background is in traditional full-stack software engineering, I am actively diving deeper into AI integrations.

When reviewing the requirements for the AI Web Developer role at Invenio Search Group, I noticed a strong emphasis on building "unattended automations that recover from failure." I decided to spend some time to actually build it.

I created this Guardian Agent to demonstrate how my foundational engineering skills (Node.js, Express, state management via SQLite) translate directly into building resilient, fault-tolerant AI systems. It proves I understand the difference between a brittle API call and a robust AI workflow.

Note: The current prompt in index.js is intentionally sabotaged with a conversational prefix to force the AI to break the JSON parse on Attempt 1, explicitly demonstrating the self-healing recovery loop in the UI.