PaperPilot

Inspiration

Whether through our own experiences or those of our friends and family, we each understand that small business owners often find themselves burdened by an abundance of paperwork that they need to draft, revise, and distribute. So, we set out to develop a solution that does all three in one, saving small business owners time, money, and stress.

What it does

We use robust AI models to streamline document creation, revision, and distribution. After prompting the user for input, we generate a document (contract, invoice, etc.) based on their description. Then, the client can seamlessly revise the document by communicating directly with our AI. When satisfied, they can email the document straight from their dashboard.

How we built it

We used Next.js on the front end and Flask and MongoDB on the back end. We built our model around the Gemini API and pretrained it using LaTeX templates scraped from the web. We also implemented a RAG architecture to improve the performance and consistency of our model in revision tasks.

Challenges we ran into

We lost several working hours to package discrepancies between our machines, the most notable of which was an outdated version of OpenSSL preventing us from accessing MongoDB.

We also had a hard time converting LaTeX into HTML that we could render and interact with on the web, all while preserving the format. We eventually settled on adapting Python and Tex4ht scripts, which preserved much of the control.

Accomplishments that we're proud of

From the start, we knew we wanted to implement many features. Despite the challenges, we are very happy with our progress and the quality we achieved.

What we learned

For all three of us, this was our first dedicated attempt at a Hackathon. We each had our own experience with code, but had never faced such a time constraint. It taught us a lot about our capabilities and also a lot about different technical frameworks that we were unfamiliar with.