Auto Helper

Inspiration

As our team wraps up our first year of university, we find ourselves in the market for cheap transportation for the coming year. A common difficulty we found was the car marketplace, of keeping track of the hundreds of different posts and listings available to you. We aimed to solve this issue, creating a tool that could scrape thousands of car listings and provide personalized recommendations.

What it does

Auto Helper is an AI assistant that uses data scraped from various car retailers in order to give personalized recommendations on cars. The users can base their search on various metrics including price, color, mileage, brand, and many others.

How we built it

We used the BeautifulSoup library in Python to build the web thousands of car listings across the web
We utilized the Pinecone database in order to store embedded listings
Langchain and Next.js was used for Retrieval Augmented Generation and to build and fine-tune a GPT model
We used React, Typescript, and Tailwind to build the frontend

Challenges we ran into

One of the biggest challenges our group ran into was the vectorization of our data. We needed to build a RAG system in order to manage data load on the GPT-model. The big struggle was figuring out how to vector embed the web scraped Json file into a database and the Langchain library's documentation proved quite hard to understand. Another challenge we ran into was trying to web scrape images. A site we were using loaded temporary images as placeholders which we found difficult to avoid scraping, but we eventually were able separate the placeholder vs the actual images.

Accomplishments that we're proud of

Our whole group is very proud of the final result. As this was many of our first times attending an in-person hackathon, we came in with pretty mild expectations. We ended up building a project that exceeded the expectations of everyone in the group and challenged us to learn new tools and techniques.

What we learned

On top of all the new technologies and libraries we used (LangChain, Beautifulsoup, Pinecone, etc.) We all agreed that the most important skill we developed our teamwork skills, especially in a development environment.

What's next for Auto Helper

We had a few features that we had to scrap for the final design of our project. Most notably, we had hoped to implement a 3d model on the site that would change depending on the vehicle. Although implementing this idea would clearly take a lot of resources, we think this demonstrates the scalability of our technology and the potential for it to grow. Another interesting option is to abstract this current program, and to make it work for not only car listings but any sort of online merchandise.