LingoGPT

A AI guided language learning app for Columbia's CodeCollab 2026.

The project is closed sourced.

Idea

Something common among our team was that we all liked learning languages and the issues common in most language learning apps. We wanted to make an app that allows users to learn language by interacting with real world scenearios. The only way to do that was to simullate events using the help of LLMs.

Architecture

As a team we spent the first couple of weeks working on the design of the app before anyone wrote anything. We settled on using a python based backend that connects to both a document based database along with a Vector database. The LLM is used upon account creation to generate user specific parameters and once again before every lesson. We settled upon a couple models that worked well with our systems and provided it with two layers of memory for both persistant and specific context. We also used prompt and parameter engineering to ensure every output syntactically works with our harness. For the front end, we used svetle. While we wanted the app to be mobile eventually, we decided that the completion of the app before the deadline was more important and opted for an easier framework.