scAInce

Asking for backpropagation
Asking for double slit experiments
Asking for a history question

ScAInce - your AI-powered study buddy

Inspiration

As students from different backgrounds and education levels, our team was united with the idea of making the learning process accessible and efficient by utilizing the technology we have. Yes, the current world of AI offers us many handy tools, one of the most widely used ones being ChatGPT. However, when it comes to visual explanations, we found out that the current AI tools lack accuracy and explanatory effects.

A study found that three days after learning new information, people recall only 10-20% of what they learned through text or voice compared to 65% of what they learned visually.

Moreover, it is widely known that people diagnosed with autism, ADHD, and dyslexia find it easier to learn with visual additions to plain text or verbal information because of the color-coding and creating visual associations.

What it does

Our AI-powered chatbot can become a powerful tool to learn and master challenging academic topics. Here is a summary of the website's functionality:

Explains the concept that is being asked step-by-step, fostering the user to build up the knowledge of the subject starting from the fundamentals.
Explains the reasoning behind every answer it provides.
Provides visual interactive graphs, charts, and other visualizations to better grasp the topic
For subjects that do not require straightforward visualizations, it provides relevant YouTube videos.
When formulas are outputted, it is verified through Wolfram Alpha.

How we built it

Created a simple website using next.js, that consists of the chatbot interface,
Defined a few types of responses the AI agent could provide us with and wrote code for that as well.
For the AI agent, we used an LLM and provided it with a list of tools such that it could make use of these tools to explain concepts to a user.
- For example, if the AI agent wants to explain a concept using web code, it would return the code that visualizes the concept. We would then take this code and run it inside a sandboxed environment.
Putting together a couple of these AI responses (math simulations, YouTube integrations) was enough to create a compelling AI tutor.

Challenges we ran into

Enabling mathematics engine to work seamlessly with the application, since it was quite cumbersome.
Coaxing the LLM to provide output that guides the user through a concept instead of providing the solution straightaway.
Limiting the possible integrations to the AI agent, to ensure good output for the user.

Accomplishments that we're proud of

Providing effective visual animations for concepts being tutored
Finding and embedding engaging YouTube videos for literature, history, geography & chemistry questions.

What we learned

Gained experience with working with LLMs
Learned how to create a responsive user interface
Learned how to implement open source tools into projects

What's next for scAInce

Add more resources for visualizations to cover more subjects
Add a quiz feature, which tests the user’s understanding and becomes gradually harder
Offline Functionality
scAInce is currently a standalone chatbot and does not integrate with platforms like Moodle, Blackboard, or Google Classroom at this stage.
Mobile App & WhatsApp Integration.