Agentic AI for Self-Healing Support (Headless E-commerce Migration)

Override Context

Hackathon Demo: Optimized for explainability, controlled autonomy, and a clear agent loop. Scenario: Self-healing support for a SaaS platform migrating merchants to headless architecture.

Problem Statement

A migration to headless architecture has caused:

• Broken checkouts
• Webhook failures
• Deprecated API usage
• Frontend-backend mismatches

Support teams are overwhelmed. This Agent steps in to Observe, Reason, Decide, and Act.

Architecture & Tech Stack

• Backend: Python (FastAPI)
• Frontend: React + Vite (Dynamic Dashboard)
• Agent components:
  • Orchestrator: State machine managing the Observe -> Reason -> Decide -> Act -> Feedback loop.
  • Observer: Ingests simulated signals (logs, webhooks).
  • Reasoner: OpenAI-powered analysis (Chain of Thought).
  • Guardrails: Risk-based permission system (High risk = Human-in-the-loop).
• Data: In-memory vector store & state for demo speed.
• Simulation: JSON-based scenarios driving the agent.

Folder Structure

self_healing/
├── src/
│   ├── agents/          # Orchestrator & Tool definitions
│   ├── engine/          # Observer, Reasoning (LLM), Guardrails
│   ├── data/            # Knowledge base & Vector store
│   └── state/           # In-memory persistence
├── simulations/         # JSON scenarios (e.g. Broken Checkout)
├── ui/                  # React Dashboard & Trace Viewer
├── main.py              # FastAPI Entry Point
└── README.md

Setup

1. Install Dependencies: pip install fastapi uvicorn openai (plus frontend deps)
2. Environment: Set OPENAI_API_KEY.
3. Run: python main.py (Starts Backend) & npm run dev (Starts Frontend).