LexiCoach

Learning language on your own terms.

Comment

Inspiration

Building vocabulary in a new language is always difficult, and sometimes flash cards, quizlets, and duolingo lessons won’t cut it.

Memorizing the translation of a foreign word is not the same as knowing that word in the context of its language.

LexiCoach leverages GPT-3 to let you to practice your own custom vocabulary sets on sentences generated in a target language.

How to use

  1. Input vocabulary that you want to be confident using.
  2. Let LexiCoach generate practice fill-in-the-blank exercises.
  3. Repeat and learn more nuanced and varied usages of your vocabulary set.
  4. Incorporated donation support with the convenient checkbooks API.

How we built it

The core of the model is prompt engineering for GPT-3's davinci LLM. The backend was built with Flask and SQLAlchemy, and the frontend with HTML and JS. The donation invoice functionality was built with the Checkbook API.

Challenges we ran into

None of us had worked with Flask or with OpenAI's API before, so we all had to learn on the fly.

Accomplishments that we're proud of

We built a viable prototype that allows you to practice a language's vocabulary in context. Crucially, it allows you to choose what kind of vocabulary you want to learn, and it introduces you to ways that this vocabulary may be used.

What we learned

We learned a lot more about what a web stack looks like and how to design effective prompts for large language models.

What's next for LexiCoach

More intentional sentence generation – deliberately evoke different meanings and inflections.

Expanded support for more languages – how to handle languages with extensive inflectional morphology

Fine tune model and eliminate problems with ambiguous answers.

Built With

Share this project:

Updates