This demo showcases an AI-driven voice assistant for vehicles, leveraging Generative AI and MongoDB Atlas to provide a smarter, more intuitive in-car experience. Unlike traditional voice assistants that are limited to basic commands, this solution enables context-aware interactions, real-time vehicle diagnostics, and personalized recommendations for drivers.
The industry is rapidly evolving with EVs, autonomous driving, and advanced safety features, driving demand for intelligent in-car assistants.
Most drivers use in-car voice assistants, but they are limited to basic tasks like navigation, calls, and simple commands.
Generative AI transforms voice assistants from basic command-response systems into dynamic, interactive copilots that:
- Understand driver needs in real time
- Provide relevant insights using vehicle data
- Enhance overall user experience
MongoDB Atlas plays a key role in this demo by providing:
- Flexible Data Storage: Easily stores and organizes operational, metadata and vector data, including vehicle signal data, car manual embeddings, and user interactions.
- Vector Search for Context-Aware Responses: Enhances voice assistant capabilities by retrieving relevant sections of the car manual based on user queries.
- Real-Time Syncing: PowerSync’s MongoDB Connector ensures vehicle signal data is always up-to-date, whether on the cloud or in the vehicle.
Learn more about MongoDB here.
- Next.js for the frontend framework
- MongoDB Atlas for the database and vector search
- Google Cloud Platform for the LLM, speech-to-text, and embeddings
- PowerSync to persist vehicle data in MongoDB Atlas and keep it in sync
Before running the demo, ensure you have the following:
- A MongoDB Atlas Cluster
- Node.js 20 or higher
- A GCP account with access to Vertex AI APIs
- A PowerSync account for real-time data syncing
To configure the environment for this demo, you need to set up the following variables in a .env.local file. These variables ensure seamless integration with MongoDB Atlas, Google Cloud Platform, and other services used in the application. Below is an example configuration:
MONGODB_URI="<your-mongodb-connection-string>"
DATABASE_NAME="car_assistant_demo"
GCP_PROJECT_ID="<your-gcp-project-id>"
GCP_LOCATION="us-central1"
VERTEXAI_COMPLETIONS_MODEL="gemini-2.0-flash-001"
VERTEXAI_EMBEDDINGS_MODEL="text-embedding-005"
VERTEXAI_API_ENDPOINT="us-central1-aiplatform.googleapis.com"
NEXT_PUBLIC_ENV="local"- Install dependencies:
npm install
- Start the development server:
npm run dev
- Open http://localhost:3000 in your browser to access the assistant and vehicle dashboard.
- Ensure you've created an
.env.localfile with valid API keys and environment variables. - Check that your MongoDB Atlas cluster is properly set up and accessible.
- Verify that your Google Cloud Vertex AI and Speech-to-Text APIs are enabled.
For a more realistic setup, you can add additional onboard capabilities by using an embedded database to hold your data and seamlessly sync it to the cloud. Since this demo highlights mostly the cloud components of the assistant, you will be able to run it without PowerSync, but we encourage you to do the full setup for a more complete view of the demo.
The application has been integrated with PowerSync for local-first capabilities and uses the default IndexedDB VFS solution. The ./src/lib/powersync directory contains all the files used for setup and talks to a separately hosted backend API for generating tokens, validating tokens and handling the upload to the source MongoDB database.
There are also two providers in the ./src/components/providers directory which perform the connection and disable SSR. PowerSync will not work with SSR at this time due to assets that need to be present on the client device.
The PowerSync SDK is used in the VehicleContext.js context to write the changes to the vehicle variable to the local database and then PowerSync will push the changes to the Backend API, which then writes it to the source backend database.
The PowerSync backend provides endpoints for token generation, public JWKS keys, and handles writes to the source MongoDB Atlas database. You can find the backend repository and full setup instructions here: genai-in-car-voice-assistant-backend.
After setting up the backend, there are two additional environment variables that need to be added to the project in order to connect successfully:
NEXT_PUBLIC_POWERSYNC_URL="<your-powersync-url>"
NEXT_PUBLIC_BACKEND_BASE_URL="<your-backend-base-url>"