AutoAssist

Inspiration

Small automobile workshops often rely on notebooks, phone calls, and disconnected records to manage service operations. Finding a vehicle's previous service history, reviewing pending jobs, or preparing an inspection checklist can take unnecessary time.

AutoAssist was inspired by the idea of making these everyday tasks easier through a conversational AI assistant.

What it does

AutoAssist is an AI-powered automobile workshop assistant that helps technicians and workshop staff access information through a simulated Alexa+-style conversational interface.

It supports three working features:

Pending Service Jobs: Retrieve current jobs, vehicle details, service requirements, and assigned technicians.
Vehicle Service History: Look up previous service visits using a vehicle registration number, including maintenance performed, mileage, and technician information.
Maintenance Checklists: Generate structured vehicle service checklists with interactive checkboxes and real-time completion tracking.

The current prototype uses synthetic workshop records to demonstrate these workflows.

How I built it

AutoAssist uses Python, FastAPI, Google Gemini, JavaScript, HTML, CSS, Docker, and Model Context Protocol (MCP).

I implemented a self-hosted MCP server using Streamable HTTP and exposed three tools:

get_pending_jobs
get_vehicle_service_history
generate_service_checklist

AutoAssist uses direct MCP tool calls for predefined quick actions, enabling faster responses. For free-form conversational requests, Google Gemini interprets user intent, selects the appropriate MCP tools, and generates responses using structured workshop data.

The system also includes input validation, error handling, and deterministic fallback responses.

I built a responsive web interface for demonstrating the simulated Alexa+-style experience and deployed the application using Docker on Render.

The project includes 72 automated tests covering application functionality, APIs, MCP communication, and security.

Challenges I ran into

A major challenge was ensuring that conversational requests triggered the correct MCP tools instead of returning generic AI-generated answers.

I also worked on making structured information easy to understand through service cards, vehicle history records, and interactive checklists.

During frontend testing, I discovered a browser-scrolling issue caused by incorrectly positioned checkbox inputs. I investigated the layout and corrected the checkbox positioning to prevent unexpected scrolling.

Another challenge was improving response times. Initially, predefined quick actions used the complete Gemini conversational workflow, which could introduce unnecessary delays.

To address this, I optimized the predefined actions to invoke MCP tools directly while preserving Gemini-based routing for manually typed questions.

Keeping the application reliable when AI services were unavailable was another challenge, which led to implementing deterministic fallback behavior.

Accomplishments that I'm proud of

Built a working MCP-powered assistant with three workshop workflows.
Integrated Google Gemini with structured workshop information through MCP.
Implemented direct MCP calls for faster predefined actions.
Created interactive maintenance checklists with completion tracking.
Developed a responsive conversational interface with structured service cards.
Deployed a working web application using Docker and Render.
Achieved 72 passing automated tests.
Created a focused, practical prototype for the Alexa+ track.

What I learned

This project helped me understand how Model Context Protocol connects conversational AI applications to tools and structured data.

I gained practical experience with FastAPI, Streamable HTTP, AI tool routing, application testing, Docker deployment, and designing reliable conversational interfaces.

I also learned the importance of balancing AI-powered flexibility with predictable tool execution and response performance.

Most importantly, I learned that a useful AI assistant should do more than generate text—it should help users complete meaningful tasks through reliable tools.

What's next for AutoAssist

Future improvements could include:

Spare-parts inventory management
Customer service reminders
Job status updates
Integration with real workshop management databases
Additional vehicle diagnostic and maintenance workflows

AutoAssist currently demonstrates a simulated Alexa+-style experience using synthetic workshop data rather than a direct integration with a physical Alexa+ device.

The goal is to explore how MCP-powered conversational tools can simplify real-world workshop operations and provide a foundation for future integrations.