correctAI

A psychology teacher was spending an entire day correcting exams.
The teacher ask for help from her students Anter and Bruno.
The students came up with the solution.
correctAI was the solution.
Now the teacher spend less than 30min to correct all the exams.

Inspiration

Seeing teachers spend hours correcting the student's exams, repeating the same task over and over has called me to action, because as a former teacher I've experienced it very well and I know I could automate this task.

What it does

CorrectAI corrects student's exam within seconds using a LLM to automate the task of reviewing and validate a student's exams according with a pre submitted exam key. If an exam key is not submitted the model will evaluate the student according with its own trained data.

How we built it

We have used HTML, CSS and JS to build the frontend. And we used technologies such as uvicorn and fastAPI to build the backend API.

We used an OCR model from Huggingface to recognize the handwritten caracters locally such as: zai-org/GLM-OCR.

Challenges we ran into

The challenge was in the handwritten exams, because the model the took too long, because we haven't the appropriate machine to run it. But we decided to use a cloud service to overcome this problem, specifically the Azure Cloud services.

Accomplishments that we're proud of

We are proud of being able to create such useful software, that we believe that will help a lot of teachers optimizing their correcting time. And we proved once again that the technology did not forget the education.

What we learned

We've learned how to build a software that is capable of handle handwriting, we learned how to use AI to validate and reviewing real world document with meaningful value.

What's next for correctAI

The next steps are deploying our solution in order to help the teacher with their correction and receive feedback about how we can improve our methods and expand our project to collaborate with other developers.