Inspiration
Bloom's 2 sigma problem is that the average student tutored one-to-one using mastery learning techniques perform two standard deviations better than students educated in a classroom environment.
The internet made content free, but it didn't solve Bloom's problem.
GenAI can by creating content and feedback based on each student.
What it does
GoLearn helps engineers learn Go. It creates code snippets for a student to read and predict its behavior through 3 multiple choice questions.
The core mechanism is that the snippets become easier or harder depending on how well a student answers questions. When a student chooses an incorrect question it identifies the concepts they likely misunderstood and makes it more likely for future questions to be related to those blind spots they have.
How we built it
We worked on a large prompt to provide Clause with significant context.
We had to set expectations for:
- What is an easy or hard question
- The concepts it had to cover
- Examples that avoided overfitting
- Feedback loop of a student's performance so that his performance influenced future questions
On the full-stack side, we used Next.js.
Challenges we ran into
- Avoiding text before and after the JSON
- We ran out of time to correctly parse the JSON into their React components
Accomplishments that we're proud of
Our final prompt is consistently returning high quality questions and choices that achieve our objective in the right format.
What we learned
Effective prompt engineering.
What's next for GoLearn
- Fixing parsing errors into the React components
- Displaying feedback on correct and incorrect answers
- Displaying information on why something is correct or incorrect.
Built With
- claude
- langchain
- next.js
Log in or sign up for Devpost to join the conversation.