Teach2U

• 

  Teach2U User Interface

• 

  Chatting with Teach2U

• 

  From Document to Questions

💡 Inspiration

The inspiration behind Teach2U is the Protégé effect, which states that students who teach others are more motivated to study the content and are able to retain and apply the content more effectively. Studies have already been conducted on the use of teachable assistants (TAs); one notable example is Betty's Brain by Vanderbilt University, which helps students learn 8th-grade Biology. However, such models are usually field-specific. With the advancement of large-language models, a potential for highly flexible and interactive TAs has surfaced.

📦 What it does

The user inserts their learning materials into the app. The app then asks the user a series of questions based on the learning materials, with the aim of helping the user to recall the information and find gaps and misconceptions in their knowledge. The conversation can then be exported as a csv file to be imported into popular flashcard applications such as Anki and Quizlet.

🛠️ How we built it

The app and chat interface was built using streamlit due to its simplicity and fast deployment time.
Text elements were extracted from the learning material using unstructured, for its ease and compatibility with langchain. The text elements are then passed to a hugging face pipeline using the t5-base-e2e-qg model within a langchain for future scalability. Due to privacy concerns, a smaller model was used so that the entire application could be run locally without a GPU.

🌋 Challenges we ran into

One challenge was the optimisation of the app. Due to the nature of streamlit, the script would be re-run on every interaction, making it resource-intensive. This problem was resolved through caching of common resources, which rendered it light enough to be hosted on Streamlit community club.

🏆 Learnings and accomplishments that we're proud of

Learning to use the langchain and transformers libraries.

What's next for Teach2U

• [ ] User response validation
• [ ] More fine-tuned, efficient models
• [ ] Export conversations as mindmaps

Built With

• huggingface
• langchain
• python