Lexi

Overview

This project was built at the first hackathon I ever attended, Hack Midwest 2023. Back then, ChatGPT was entering the mainstream, image generation models were getting better, and people were just starting to build GPT wrappers at hackathons. Our idea for this one was to combine LLMs and Stable Diffusion image generation to create custom targeted storybooks for learning languages. Users can enter vocab words, and the app generates a story containing those words, with pictures accompanying each sentence.

Technical Details

We used:

• GPT 3.5 Turbo for generating storylines in Spanish based on a list of vocab words
• Stable Diffusion for generating a cartoony picture for each sentence
• DeepL for translating the story back to English (so the user could flip between languages)
• The Pinata API to persist the storybooks to IPFS, so that they could be reused anywhere on the web with our dedicated gateway

Hack Midwest 2023

By some miracle, we won the prize for the Pinata Challenge!

Some fun details about this whole experience:

• Our team lead who convinced us to attend the hackathon in the first place ( Blaine) got sick that day and couldn’t make it, so we pulled in a sub last minute
• He was the only one with experience in server side JavaScript, so I had to learn Express.js and some other stuff (this was before the age of vibe coding!)
• First time doing literally anything with AI
• In the end the app would break half the time, we demoed twice and both times we lucked out and it worked lol
• This hackathon win actually got me a summer internship at Pinata!

Repository

looking at this now two years later this code looks like complete garbage ngl

Last updated on December 28, 2025.