GrokLearn

Inspiration

When I was using Grok voice to learn some new things this year (AI ML concepts & 1900s history), it was useful, but a few days later I failed to recall many of the things I learned. I wanted a system that would collect flashcards as I conversed and could quiz me on them later with the ability to full navigate and resume previous topics via voice.

What it does

There are two main modes: learn & review. The learn mode recalls the most recent knowledge you had for a topic based on previously created flashcards and continues the conversation while automatically creating flashcards for you based on what you discuss with the assistant. The review mode selects flashcards you previously created from the database, quizzes you, validates your answer, and explains to the correct answer (and why) if you got it incorrect.

How we built it

Implemented a multi-agent router and scaffold for the two different agents with system prompts and relevant data loading to make the experience feel customized. We built this with typescript + react.

Challenges we ran into

At first we started off using the WebRTC API but there were issues with streaming, so in order to implement the agent router we had to switch to the normal Web API which sends a session key to client, but this requires starting a new session on the

Accomplishments that we're proud of

The agent router, automatic flashcard creation, and review mode work extremely well. This is something we will use!

What we learned

Agent routers work very well and far more complex agent routing setups would work, and could allow us to create a full Jarvis-like agent!

What's next for GrokLearn

We'd love to come build this at xAI!

Built With

• grok
• react
• typescript