Nutshell

Inspiration

Nutshell came about out of necessity. All members of our team are addicted to Facebook Marketplace, and we're all cheap as hell. Of course, one of our favorite shared pastimes being cheap as hell, is lowballing people on Facebook Marketplace. However, our offers don’t often go over well, whether that be stingy sellers or our insulting offers. This is why we decided to make a browser extension that would use cutting-edge AI technology to get us the best deals with no effort.

What it does

Nutshell, in a nutshell (see what we did there?) analyzes the listing info, such as price, features, and the listing images to build an offer index, that tells you if you need to get off that listing page, send in an offer (lowball or fair) or if you need to jump on this deal before someone else gets it.

How we built it

We used Python with PyCharm, Claude, and VS Code for the Chrome extension.

Challenges we ran into

We ran into a few challenges, especially with the fine-tuning. But after lots of debugging, we eventually made it.

Accomplishments that we're proud of

• Finetuning 2 LLMs using methods such as LoRA finetuning
• Implementing everything into a simple API that is accessed by a clean Chrome Extension

What we learned

• We learned how to use MCP agents and Claude Code
• We learned how to fine-tune DistilBert (normal LM) and Paligemma2 (vision LM)
• We learned a lot about hackathons in general. Mostly that we need more coffee.

What's next for Nutshell

Polymarket/prediction market implementation

Built With

• fastapi
• pandas
• python
• transformers