TeachSim

Have you ever wanted to improve your coaching skills? With TeachSim, you can practice explaining in front of a group of AI students that consider your explanation and ask probing questions.

Comment

Inspiration

We were inspired to build TeachSim from brainstorming ways to leverage GenAI to help develop real human skills. It was an interesting concept to us to combine similar inputs/interactions from many different models to simulate a real-world classroom and be able to have the human respond to AI's prompts for a change.

What it does

You teach a group of AIs a topic, and they ask questions and you receive feedback on how well you accomplished teaching. The user gets probed on different perspectives to their lesson and gets to demonstrate their knowledge and learn how to explain in a gameified UI.

How we built it

Using a Flask (in python) backend, we interfaced with a HTML/CSS/JS frontend and OpenRouter to send API calls to different AI models, receive generated JSON responses, treat them with some simple algorithms (sometimes also AI assisted) and display information on the user's web browser.

Challenges we ran into

It was our first hackathon and first time using AI APIs for all members on the Floor 5 team, and this brought a widge range of challenges to the team.

  • Brainstorming: We spent a large amount of time deliberating on an app idea that incorporated AI and allowed us to expand our toolkits.
  • Synchronized workflows: Many of us had to adapt to using Git in new ways (and for the first time, for some of us)
  • Problem solving: For many issues, the best methodology was unclear early in the development process and we often found ourselves considering several solutions to a problem, choosing one, and then backtracking later due to unforeseen factors. For example, the Gemini API allowed for some great JSON schema validation, but we quickly ran into rate limits on our personal API keys.
  • Documentation: We weren't very good at journalling our brainstorming or thought processes which sometimes made it hard to return to earlier trains of thought or following each others' logic trains. It would've been helpful to form an abstract mission and objectives first without considering a method, and then adapting that later.
  • Research: As a lot of this territory was new, we found ourselves doing a lot of research and reading docs instead of designing.

Accomplishments that we're proud of

  • We managed to coordinate well on a final product and each contributed our skills -- while still learning a little bit about every part of the process
  • The app successfully queries AIs in a reasonably efficient way, not spamming tokens too often and making use of single large calls rather than frequent small ones.

What we learned

We learned a lot about frontend/backend communication, API usage, HTML/CSS styling, Python libraries, and most of all, LLM models. We have a much greater understanding of LLMs' place in development, the competitive landscape, and how to use and store API keys.

What's next for TeachSim

We're hoping to build a nicer UI and expand with some more "homely" features like maybe a voice-lecture mode or smoother animations.

Built With

Share this project:

Updates

Submission history