LLM Workflow Automation Engine

A production-style proof‑of‑concept demonstrating LLM‑powered ticket triage, workflow orchestration with n8n, and a clean FastAPI + HTML/JS frontend. Built to showcase real‑world automation architecture for employers evaluating AI engineering capability.

Why This Project

This repo is designed as a realistic, production-style example for employers evaluating AI workflow engineering skills. It demonstrates:

End-to-end architecture (frontend → API → LLM → workflow engine)

Structured JSON outputs from an LLM (ticket summary, category, priority, review flag)

Integration with an orchestration tool (n8n webhook → FastAPI) suitable for real-world automation.

---

🌐 Overview

This project implements an end‑to‑end workflow:

1. Frontend (HTML/JS) – A simple Ticket Triage Console where a user submits:

   • customer name
   • email
   • ticket text

2. Backend (FastAPI) – Receives the ticket and runs the core LLM classification:

   • summarizes the issue
   • assigns category
   • determines priority
   • flags whether human review is needed
   • returns structured JSON

3. Workflow Orchestration (n8n) – Optional automation layer:

   • Production webhook endpoint
   • HTTP Request → FastAPI
   • Webhook Response returns structured result

---

🏗️ Architecture

[ Frontend ] → [ FastAPI ] → [ LLM Engine ]
↑
|
[ n8n Webhook ]

---

📁 Project Structure

llm-workflow-automation-engine/
├── src/
│   ├── api_server.py
│   ├── workflow_engine.py
│   ├── prompts.py
│   ├── config.py
│   ├── schema.py
│
├── static/
│   └── index.html
│
├── n8n/
│   └── ticket-intake-llm.workflow.json
│
├── requirements.txt
├── config.example.toml
└── README.md

Reliability & Validation

All LLM responses are coerced into a structured schema (see schema.py).

The workflow engine validates that required fields are present before returning a response.

Errors are surfaced as HTTP 4xx/5xx with JSON error bodies for easy monitoring and integration.

---

🔧 Setup

1. Clone the repo

git clone <your_repo>

2. Create venv

python -m venv .venv

Windows

.venv\Scripts\activate

macOS/Linux

source .venv/bin/activate

3. Install deps

pip install -r requirements.txt

4. Copy config

copy config.example.toml config.toml

Fill in API key + model.

---

▶️ Run FastAPI + Frontend

uvicorn src.api_server:app --reload
→ Visit: http://127.0.0.1:8000/

---

🌐 n8n Setup (Optional)

Run n8n: n8n

Import workflow file in n8n/ticket-intake-llm.workflow.json.

Production URL will look like: http://localhost:5678/webhook/ticket-intake-llm

---

📈 Sample Output

summary: Customer reports a double charge
metadata:
customer_name: Jane Doe
email: jane@example.com
priority: high
category: billing
needs_human_review: true

---