代理-AUTOMOTOMOT-SUP-SUSS助剂

Agentic AI Automotive Support Assistant

An AI prototype that proactively identifies and resolves vehicle issues before a support case is created. This system uses a modular, MCP-compliant architecture, multi-agent coordination, and LLM-based reasoning to reduce customer friction and support load.

✨ Overview

When a customer taps "Help" in their mobile app or in-car head unit, this system launches a series of diagnostics and knowledge base lookups via multiple MCP servers. A reasoning agent then synthesizes these signals and either resolves the issue or escalates it with full context—without needing human intervention.

This prototype demonstrates:

• Agentic workflows with crew.ai
• Prompt evaluation with promptfoo
• Reasoning pipelines with LangChain
• Infrastructure-as-Code with AWS CDK (TypeScript)
• Orchestration using Model Context Protocol (MCP)

💡 Use Case

"Remote Start isn’t working" — before the user submits this as a support ticket, the assistant:

1. Inspects recent user actions ("user tapped Remote Start 2x")
2. Checks car configuration ("feature not yet enabled by user")
3. Verifies subscription status ("remote services expired last week")
4. Looks for outages ("known issue impacting 5G connectivity in US West region")
5. Retrieves similar issues from knowledge base (via bedrock-kb-prototype)
6. Summarizes a response to the user and recommends actions, or solves it.

🤖 Architecture

[ Mobile App / Head Unit (MCP Host) ]
             |
        [ MCP Client ]
             |
   -----------------------------
   |     |        |           |
[Config][Subscription][Outage][KnowledgeBase] ⇨ (MCP Servers)
                         |
                 [ crew.ai Agent Team ]
                         |
                [ LangChain Planner Agent ]
                         |
                  [ LLM Summary to User ]
                         |
                [ promptfoo Prompt Evaluation ]

🚀 Stack

🔧 Setup

Coming soon: Full CDK + deployment instructions.

You can run LangChain logic + crew.ai planner locally and mock API Gateway inputs to simulate the MCP payload.

General Tutorial for Building Similar Projects:

1. create a fresh GitHub repo
2. create Amplify Gen 2 project from within the AWS Console and point it to fresh repo
3. from CLI, run npm create amplify@latest in repo's root folder.
4. git push and check Amplify console to ensure that deployment worked.

🔹 Appendix: Ambient Computing Extension

Future evolution of this project will include passive telemetry + usage monitoring to detect and solve vehicle issues before the customer even opens the support app. This aligns with the concept of ambient computing and predictive support UX.