MaThinq

• 

  MaThinq in action!

• 

  Website Overview!

Inspiration

The math performance of American students has declined over the last 25 years. Research shows that students learn differently, and in today's day and age, more and more children prefer to learn from digital content. While digital content and teachers exist, there is no existing generative software that can specifically create videos related to a student's particular needs.

What it does

Users can enter a math related query into our free software. MaThinq will understand the query, and generate a custom video with audio to help the student learn. Additionally, MaThinq generates a practice problem relating to the user's query.

How we built it

Backend: Python, FastAPI, Manim, OpenAI, SQLite, LaTex, Matplotlib, Pandas Frontend: React.js

Challenges we ran into

There were several big challenges we ran into in this project. The biggest challenge was ensuring high quality videos and avoiding objects leaving the frame, mathematical inaccuracies, overlapping text, latex rendering issues, etc. Another challenge was managing latency in both the generation of content and the rendering of manim code. We managed this by tuning hyperparameters of both the AI generations and aspects of the videos (frame rate, video quality, etc.)

Accomplishments that we're proud of

We're particularly proud of Mathinq's ability to generate high quality, mathematically accurate videos. The state of the art in video generation (diffusion models, SORA, etc.) is good at creating visuals, but is unable to guarantee accuracy, mathematical correctness, or consistency. By grounding our video generations in mathematical reason and the use of Manim, we're able to consistently create videos that are mathematically accurate and informative.

What we learned

We learned about machine learning (prompt engineering, model selection, parameter tuning, etc.) and also about the software side of creating an application (managing latency, ensuring consistency of results, frontend, UI/UX design, usability, etc.).

What's next for MaThinq

We aim to fine-tune a model in Manim generation to reduce any potential overlapping content, mathematical inaccuracies, and other common issues with rendering.

We'd also like to partner with educators or e-learning experts to integrate this tool with existing learning methods or sources.

Discord: Isaac Tu: isaactu. Rahul Narayanan: indiot2004

Built With

• fastapi
• latex
• manim
• matplotlib
• openai
• pandas
• python
• react
• sqlite