Teach me anything robot

Inspiration

We aimed to apply pre-trained large-scale language model to build an education chatbot. Apart from proper prompt searching and testing to find the correct prompt for the desired output, we aim to improve one of the fundamental problem in the LLM - a limited context window. We implemented a hierarchical summarisation model to solve this problem.

What it does

• Users provide a topic
• The language model creates a list of subtopics
• The user picks a subtopic and the model teaches them about it
• The conversation is heirarchically summarised, so that the model can remember everything you've talked about (even if it's an... infinite amount)

How we built it

• Frontend: HTML, CSS, JavaScript
• Backend: Flask, LM Studio

Challenges we ran into

• Best prompt searching
• New tech learning
• Llamas are really stupid

Accomplishments that we're proud of

• Finish the project
• Not too seriously ill even though not sleeping
• Not seen any hair fall out

What we learned

• LLM and how it works
• Flask framework
• Web dev

What's next for Teach me anything robot

• refinement would include a quiz testing feature

Built With

• css3
• flask
• html5
• javascript
• llama
• lm-studio
• python