educat

Landing Page
File Uploading
Learning with Teacher Cat
Student Cat Testing Your Mastery

| Inspiration

How can AI do more than just give you the answers to a homework problem? What can students do to fully demostrate mastery in learning? With LLM-powered cat tutors, one teacher and one student, we explore how a user can learn first in a bottom-up, algorithm-driven approach with an AI teacher, then show their understanding by solving the final steps themselves and tutoring an AI student!

| What it does

educat explores the ways AI can bring something unique to education. Any human can explain the answer a basic algebra problem, but only our AI cats can deconstruct a problem into a standard learning algorithm and stop just when they know the student is equipped with the knowledge to get the final solution. We've built a personalized learning experience, complete with a fully interactive user interface and fun cat-themed graphics! Users can get started easily with seamless and secure account authentication, then upload a .pdf, .txt, or .doc of a homework question or problem set to get started on a study session with our two AI-powered cats. Our learn-then-teach approach works by:

Teacher Cat breaks down a problem into its foundational subtopics and identifies prerequisites to solving the problem, creating an internal tree structure with topic nodes and directed edges. Topics also live in a user's own mind map storing mastery of content from all study sessions.

In a bottom-up approach, Teacher Cat guides understanding in the foundational requirements Socratic questioning-style. Correct answers help slowly master the user's mind map and progress through the topic stack, but when a gap in understanding is identified, Teacher Cat recursively finds the underlying cause - an earlier topic not truly mastered, and updates both the current lesson and mindmap on all overlooked topics.

Once the prerequisite stack is cleared, students are introduced to Student Cat. Users now have all the knowledge they need to solve the problem, and should now be able to explain both the solution and reasoning to Student Cat until full mastery is proven!

After a study session, the user's progress on learning concepts, visualized by a mind map, is updated to reflect their newfound understandings so that next time, our tutor will know exactly what the user already knows!

| How we built it

Secure authentication and sign up with account customization [MongoDB, React]

Interactive, user-friendly interface with hand-drawn cats [React, Typescript, Procreate]

AI-powered document ingestion and content analysis [GeminiAPI ]

"Socratic scaffolding" algorithm: A breadth-first search-like algorithm to cover prerequisites and a ?recursive search for foundational weaknesses in subtopics [GeminiAPI, ElevenLabs, Typescript]

Mastery test using an AI student to judge thoroughness and accuracy of explanation [GeminiAPI]

Persistent user context on demonstrated mastery in progress subtopics [MongoDB]

| Challenges we ran into

Some of our biggest challenges included fine-tuning how the AI broke down topics, considering factors like making sure synonym topics weren't being double counted in the mind map or making sure the granularity of a topic matched the problem itself (ex: we don't want to be quizzing the user on how addition works if they're on a quadratic formula problem). Additionally, we needed to integrate a seamless evaluation of the user's response and performance, determining what really counted as "mastering a topic", to motivate the follow up response and trigger graphical changes on the user side.

| Accomplishments that we're proud of

We're proud to have integrated GeminiAPI into a project that brings more capability to the LLMs we use everyday. Anyone can easily prompt Gemini for a homework solution, but we've broken down learning into a formulaic approach that augments AI with a human view on how learning really works. With fun yet effective study sessions, we've proven that AI can flip traditional pedagogical classroom into an experience that feels more like a Socratic personal trainer.

| What we learned

As a majority first-time hacker team, this was our first time building a full-stack web app together. We learned how to ideate from the ground-up, manage a codebase together, and tried out new tools like Typescript, GeminiAPI, and ElevenLabs! We each explored parts of the project spanning our full stack, learning from each other on responsive web design, API integration, and building something novel.