CarMommy

Never haggle for a car again.

Comment

Inspiration

One of our team members recently went to purchase a new car, and ran into a problem that nearly everyone faces: conflicting and constantly shifting prices from whatever dealership they went to visit. This caused him to have to waste countless hours negotiating and attempting to talk the price down, before finally settling on a reasonable price. After learning about this ordeal, we realized how much easier and fairer car-buying could become just by using AI to help in the process!

What it does

Negotiating prices is something that nearly everyone dreads. That's where CarMommy steps in. CarMommy is an AI-powered car buying agent that will work on behalf of a user to negotiate fair prices across multiple dealerships.

CarMommy searches through thousands of car listings by scraping Carfax to find the best offers in dealerships near you. It dispatches AI voice agents (powered by ElevenLabs) to call car dealerships and handle all the tedious number crunching (listing price, MSRP, etc) so you don't have to do the tedious work yourself. Additionally, it generates cinematic showcase videos of selected vehicles using Veo 3.1, Google’s cutting-edge video generation model, automatically selecting the best exterior photos and creating professional presentations Lastly, it also processes secure payments through the Solana Blockchain.

All of this happens automatically as users sit back and watch the best deals come to them!

How we built it

We built CarMommy using a fast, scalable stack: we used React, Typescript, and Vite for the front-end, and Convex for our backend to handle server logic and data. For the UI and design, we relied on Tailwind and Shad/CN UI to make it clean and responsive. For the voice agent capabilities, we used the Elevenlabs Agents Platform to enable automated phone calls to dealerships, and integrated Gemini API to access Google’s powerful generative AI capabilities (Veo 3.1) to make AI-created videos of cars and also used it for email post processing (Gemini 2.5 Flash)! Lastly, we also used Solana to handle secure payments and made a web scraper to scrape Carfax for real-time car data to display on the website.

Diagram

Untitled-2025-11-15-2043

Challenges we ran into

Creating a conversational AI voice agent that could naturally negotiate with real car dealerships was our most challenging task. We needed an AI agent that could hit multiple precise and complicated requirements. We needed the agent to understand nuanced car-buying conversations, handle pushback from the salesperson, extract pricing information from free-form responses, and maintain a professional yet friendly tone. This required extensive prompt engineering and testing with the ElevenLabs API. We went through dozens of iterations to find the right balance: our end goal was an AI Agent that would be supportive like a "car mommy" should be, without being too pushy or overwhelming such that they would fail to close deals in the first place. The agent needed to know when to push for better deals, when to accept reasonable offers, and how to gracefully handle rejection or unclear responses from dealership staff. Additionally, parsing the varied formats of dealer responses and extracting confirmed prices from unstructured email text required sophisticated AI prompting with Gemini and utilization of Gemini’s structured output feature to ensure accuracy and uniformity.

One of our biggest hurdles was that there were no existing APIs we could build off of to ingest real-time car listing information. We had to develop our own custom scraping solution from scratch. We built a robust web scraper to aggregate car listings from Carfax’s website to tackle this challenge. Additionally, in order to maintain accuracy, we implemented extensive error handling to deal with complicated HTML structures.

Processing thousands of car listings, coordinating multi-step AI workflows, and handling long-running async operations initially caused significant performance issues. We worked to improve performance by carefully designing our database queries to handle real-time updates as AI agents completed dealership calls, while ensuring the UI remained responsive for the users to interact with other features on CarMommy. We also had to optimize image processing pipelines, encoding images to base64 multiple times for different AI models in order to avoid performance bottlenecks or memory issues.

Accomplishments that we're proud of

Something we’re proud of is that we were able to integrate a lot of tracks and challenges into this project: using the voice agent technologies from ElevenLabs, using Solana for payments, and using Gemini API for video generation and email post-processing. Also, we’re just generally proud of the overall project that we made, since this has the potential for great social impact, as many people could benefit from this!

What we learned

All of us were completely new to Solana, so integrating it into our project was completely foreign at first, but eventually we were able to! Additionally, this was one of our team member’s first hackathon, so he learned a bunch about the entire app development process, like TypeScript, React, front-end design, Git, etc. Lastly, all of us learned a lot about how to use the agent technologies from Elevenlabs, which are genuinely so cool!

What's next for CarMommy

After the hackathon, there are several possible venues that we could expand into, but the most likely is to just flesh out the whole project and make it feel like a true personal car-buying assistant that you can depend on! Additionally, some features that we are thinking of adding are capabilities to track price changes over time and to be able to alert users when great deals pop up.

Built With

Share this project:

Updates