Module 6: Hands-on Lab: Building Your First Agent

This module is a practical, step-by-step guide to building your first AI agent in the raiaAI Launch Pad. We will apply the concepts you've learned in the previous modules to create a simple but functional customer support agent.

Objective: To build a basic support agent that can answer questions from a small knowledge base and escalate to a human when necessary.

Materials:

• Access to the raiaAI Launch Pad

• A sample knowledge base file (provided)

• The instruction template from the raia training materials

Step 1: Create a New Agent

1. Log in to the raiaAI Launch Pad.

2. Click on the "Create Agent" button.

3. Name your agent according to the raiaLabs naming convention: "Support Agent - Training - [Your Name]".

4. Select GPT-4o as the model.

5. Click "Create".

Step 2: Upload the Knowledge Base

1. Navigate to the "Training" tab of your new agent.

2. Under "Knowledge Base", click "Upload Files".

3. Upload the provided sample knowledge base file (training_faq.md).

4. Wait for the file to be processed and indexed by the vector store.

Step 3: Create the Instructions

1. Navigate to the "Instructions" tab.

2. We will use Manus to help us write the instructions. Open a new Manus session and use the following prompt:

Step 4: Review and Paste the Instructions

1. Review the instructions generated by Manus. Check that all the requirements from the prompt have been met.

2. Copy the markdown-formatted instructions.

3. Paste them into the "Instructions" field in the raiaAI Launch Pad.

Step 5: Test Your Agent

1. Navigate to the "Test" tab.

2. Ask your agent questions based on the training_faq.md file. For example:

• "What is the return policy?"

• "How much does the SmartWidget cost?"

• "My SmartWidget won't turn on. What should I do?"

1. Verify that the agent is providing accurate answers and citing the knowledge base file.

Step 6: Test the Escalation Procedure

1. Now, test the escalation rules you defined in the instructions. For example:

• "I want to speak to a human."

• "I am extremely frustrated with your service."

1. Verify that the agent responds with the correct escalation message.

Congratulations! You have successfully built and tested your first AI agent. This hands-on experience is the foundation upon which you will build more complex and sophisticated agents for our clients. In the next module, we will explore some advanced techniques for prompt engineering.