Lab guide for vehicle maintenance use case¶

Overview¶

This LAB is written to simulate a vehicle troubleshooting scenario. Here the scenario is that a vehicle that has Internet of Things (IoT) has broken down and the user is seeking for help. The car has some weird noise and the user asks the chatbot why is there a noise what does it mean? and the AI Agent goes through reasoning to find out what went wrong and provide an analysis report and find nearest service center.

Pre-requisites¶

• Make sure you've already setup the environment:
• Lab 0 - Environment setup
• ADK Installation
• Download files
• Download the zip file from Lab1 folder

Reference Architecture¶

Key Components¶

• Telematics data analyzer agent (External Agent) – Get the car telematics data and analyze it to provide a natural language summary
• Get Telematics data (Tool) – Get the telematics data of a car. (simulation)
• Get nearest service center (Tool) – Get the nearest service center. (takes lat & lon and gives results)
• Troubleshoot agent (Native wxO Agent) – A supervisor agent that orchestrate between the following:
  1. Knowledge (RAG):
     1. DTC code list.pdf
     2. Car user manual.pdf
  2. Toolset:
     1. Agents:
        1. Telematics data analyzer agent
     2. Tools:
        1. Get nearest service center

Steps¶

1. Deploy the toolsets¶

• To get started with the Agent Builder on watsonx Orchestrate UI, you will have to deploy the toolsets first.

• From the root of the downloaded files, run the following commands to deploy the toolsets:

  orchestrate tools import -k python -f "tools/get_nearest_service_center.py" -r "tools/requirements.txt"
  

  orchestrate tools import -k python -f "tools/get_vehicle_telematics.py" -r "tools/requirements.txt"
  

  orchestrate knowledge-bases import -f knowledge_bases/knowledge.yaml
  

  orchestrate agents import -f agents/vehicle_telematics.yaml
  

• Now that the toolsets are deployed, you can proceed with the agent development from the Agent Builder.

2. Create the agent¶

• Open the watsonx Orchestrate UI. Click on Create new agent on the bottom left.

  

• Select the following:

  • Create from scratch
  • Name: Vehicle maintenance Agent
  • Description: A helpful and knowledgeable vehicle troubleshooting assistant.

• Create the agent.

  

3. Configure the agent¶

• The agent will have 4 sections:
  • Profile: Describes your agent's purpose.
  • Knowledge: Provides the specific knowledge your agent needs to accomplish its goal. (RAG)
  • Toolset: Add tools and other agents to help your agent take action.
    • Tools: Gives your agent access to your organization's tool to help it accomplish tasks.
    • Agents: Identifies agents to whom your agent can delegate tasks.
  • Behavior: Defines how and where your agent should react to requests and respond to users. (System Prompt)

Profile¶

• This is the description of your agent, which you have already defined.

  

Knowledge¶

• This is the additional context that you can provide to your agent for Retrieval Augmented Generation (RAG).

  Note

  Every agent developed on watsonx Orchestrate will support Retrieval Augmented Generation (RAG) out of the box!

• For the Vehicle maintenance agent, we'll upload some documents such as vehicle's user manual and diagnostic trouble codes (DTC). We have selected publicly available data such as Jeep Compass user manual and DTC codes list.

• Add a description and upload the files from the downloaded knowledge_bases/ directory:

  • Description: This knowledge addresses the car manual, dtc codes related information to the Jeep Compass 2024.
  • Upload files from the downloaded files knowledge_bases/ directory:
    • Jeep-Compass-OM-24.pdf
    • dtc_list.pdf

  Info

  You can also connect to external repository to add knowledge base. Currently supported external repositories are:

  • Milvus
  • Elastic search

  

Toolset¶

Tools¶

• You can add tools from any of the following methods:

  1. Add from catalog: Choose from a catalog of existing tools.
  2. Add from local instance: Choose from a host of tools available on this instance.
  3. Import: Import an external tool.
  4. Create a new flow: Use the tool builder to create a sequence of steps.

• We will be using the Add from local instance. Click on Add tool and select Add from local instance.

• Select the tool:

  • get_nearest_service_center

Agents¶

• You can add a new agent from any of the following methods:

  1. Add from catalog: Choose from a catalog of pre-built agents.
  2. Add from local instance: Choose from a list of agents available on this instance.
  3. Import: Import and register an external agent.

• We will be using the Add from local instance. Click on Add agent and select Add from local instance.

• Select the agent:

  • vehicle_telematics_agent

Behavior¶

• Add the following system prompt:

You are an expert Vehicle maintenance assistant. Your job is to help car owners diagnose and understand vehicle issues by analyzing telematics data and referring to the official user manual. VIN Number: 1HGBH41JXMN109186

You are working with a vehicle_telematics_agent. This agent can get you a summary of vehicle telematics data.

You are also working with a tool get_nearest_service_center that takes lat & lon to give a list of service centers.

4. Deploy the agent¶

• Click on the top right Deploy button to make the agent live.

  

5. Test the agent¶

• Go to the Agent chat page.

• Select the Vehicle maintenance agent.

  

• Now you can chat with the agent with the suggested script below.

Suggested script¶

1. What does the engine temperature warning light mean? -> Troubleshoot agent will use the Knowledge base to RAG and answer this query

   

2. My car is shaking and I have the engine temperature warning light on can you diagnose it? -> Troubleshoot agent will transfer the control to Telematics data analyzer agent which will ask follow-up questions if required and give a car health report. The car report is read, and a suggestion is provided by the Troubleshoot agent.

   

3. Where is the nearest service center? -> Troubleshoot agent will invoke the Get nearest service center tool and pass the lat & long received from the Telematics data analyzer agent (assumption is that car will send the current lat & long data as part of telematics data.) and get the nearest service centers from the list of service centers.