Levi's Word Leap

Inspiration

Words are the primary medium through which we express ourselves. It's clear that many people are interested in expanding their familiarity with new words; after all, that is why millions of people play Wordle and hundreds of thousands also follow "Word of the Day" newsletters. We must continue to develop methods that help people build their vocabulary, for it helps people communicate effectively and descriptively.

While the aforementioned online word games are effective tools to supporting people's ability to access vocabulary resources, tools which suggest new terms in a personalized and at a higher cadence would help people even more.

"Levi's Word Leap" has people converse with an AI in a gamified setting. The vocabulary suggestions are derived from the person's conversation, thus naturally making these suggestions tailored to each individual. People can talk to Levi at any time, and, as an added benefit, they can obtain vocabulary suggestions in the context of real conversations.

What it does

Levi's Word Leap has two parts. First, is an AI-driven conversation. A person can chat with the AI, and the AI will respond in an engaging manner. Second, Levi analyzes this conversation against an English dictionary containing 100,000+ words, and identifies key terms that the person can use without significantly changing the meaning of their statements.

Levi's Word Leap gives each conversation a score, thus allowing friends to compete with each other, a motivating factor for many.

How we built it

We use Together.AI's embeddings to embed an English dictionary into a ChromaDB database. We use the Mixtral LLM to generate conversations and analyze conversations for suggested vocabulary terms.

The frontend is built using Reflex and the backend is implemented in Flask.

Challenges we ran into

It was very difficult implementing retrieval augmented generation on such a large corpus. We also faced issues with the prompt engineering.

Accomplishments that we're proud of

Our app works!

What we learned

RAG, AI, LLMs, Reflex

What's next for Levi's Word Leap

Voice chat!

Built With

• openai
• reflex
• together