Multi-Agent Customer Support Automation

Built with: CrewAI · OpenAI GPT · Python
Author: Pradeep Kumar

---

What This Does

A two-agent customer support system built with CrewAI. Instead of a single LLM handling support end-to-end, two specialised agents collaborate:

• Support Agent — researches the inquiry by scraping official documentation and drafts a thorough response
• QA Agent — reviews the draft for accuracy, completeness, and tone before delivery

The pattern mirrors how real support teams operate: a first responder handles the inquiry, a senior reviewer ensures quality before the response reaches the customer.

---

System Architecture

Customer Inquiry
      │
      ▼
┌─────────────────────────┐
│   Support Agent          │  role: Senior Support Representative
│   ─────────────────────  │  tools: ScrapeWebsiteTool (docs)
│   • Reads inquiry        │  allow_delegation: False
│   • Scrapes docs         │  memory: shared via Crew
│   • Drafts response      │
└───────────┬─────────────┘
            │  passes draft
            ▼
┌─────────────────────────┐
│   QA Agent               │  role: Support QA Specialist
│   ─────────────────────  │  tools: none (review only)
│   • Reviews draft        │  allow_delegation: True
│   • Checks accuracy      │  memory: shared via Crew
│   • Finalises tone       │
└───────────┬─────────────┘
            │
            ▼
    Final Customer Response

---

Key Concepts Demonstrated

Concept	Implementation
Role Playing	Each agent has a distinct role, goal, and backstory
Focus	Agents are prompted to stay in character and avoid assumptions
Tool Use	Support Agent uses ScrapeWebsiteTool to ground answers in official docs
Cooperation	QA Agent can delegate tasks back to the Support Agent if needed
Guardrails	Task expected_output constrains response scope and format
Memory	memory=True on the Crew enables agents to share context across tasks

---

Tech Stack

• CrewAI — multi-agent orchestration
• crewai-tools — ScrapeWebsiteTool
• OpenAI GPT-3.5-turbo — LLM backbone
• python-dotenv — secure API key loading

---

How to Run

1. Clone the repo

git clone https://github.com/Pradeep-Kumar25th/multi-agent-customer-support.git

2. Install dependencies

pip install -r requirements.txt

3. Set up your API key

cp .env.example .env
# Edit .env and add your OPENAI_API_KEY

4. Launch the notebook

jupyter notebook multi_agent_customer_support.ipynb

Run all cells. The crew will execute and print the final QA-reviewed response.

---

Customising the Inquiry

The crew accepts dynamic inputs — change customer, person, and inquiry in the inputs dict at the bottom of the notebook to test different scenarios:

inputs = {
    "customer": "Acme Corp",
    "person": "Jane Smith",
    "inquiry": "How do I configure agent memory with a custom embedding model?"
}

---

Project Structure

multi-agent-customer-support/
├── multi_agent_customer_support.ipynb   # Main notebook
├── requirements.txt                      # Dependencies
├── .env.example                          # API key template
├── .gitignore                            # Keeps secrets out of git
└── README.md                             # This file

---

License

MIT — feel free to use and adapt.