Treat Me To A Toyota

Home Page
Home Page With Search
Car Listings
Discovery Page
Listing Page
Payment Form

Inspiration

The project was inspired by the idea that users shouldn’t have to carefully type out every preference or make a series of rigid choices. We wanted to make it easier to express what you want naturally, creating a fluid, intuitive experience. At the same time, we focused on responsiveness, ensuring that the system actively acknowledges and adapts to user input as the conversation unfolds.

What it does

We’ve built a natural, human-voiced AI assistant that listens as you describe your ideal car. It asks smart, relevant questions to help refine your search, and as you narrow your preferences, a dynamic carousel updates in real time to showcase the features your AI has identified so far.

How we built it

The system leverages Claude Opus 4.5 alongside other models, using tool-calling capabilities to dynamically query and interact with the database, creating a robust, reliable, and extensible architecture. For voice interactions, we use the Vapi human-like voice agent to respond naturally and ask relevant questions. Additionally, we crafted carefully designed prompts and safeguards to ensure the system remains accurate, resilient, and safe throughout user interactions.

Challenges we ran into

The biggest challenge was selecting the right text-to-speech solution and tools to generate accurate queries for the PostgreSQL database, which required a lot of trial and error.
Another challenge was how to present car images in a way that felt responsive. For example, if a user requested a white leather seat, how could we show that effectively? We decided to use a unique carousel design that alternates between car images, allowing us to highlight all the features the user wants in their ideal car.

Accomplishments that we're proud of

The carousel created a highly engaging user experience and looked visually striking. We’re especially proud of integrating the voice agent to dynamically load car images into the carousel, making the interaction feel seamless and interactive.

What we learned

We gained valuable experience with tooling and designing effective prompts for the LLM to generate accurate SQL queries for our database. We also learned how to create a visually appealing user experience while carefully considering how users would perceive and interact with the page.

What's next for Treat Me To A Toyota

We plan to explore ways to offer more personalized pricing plans. Additionally, we aim to automate contacting dealerships to make the process faster and easier for users.