MarkSwift | Devpost

Communicating and grading the document via chat for structured output
exam paper grading

Use Case Title:

Grade your student's test papers within minutes

Description:

I used marvinAI with openAI to build this, marvin plays a role in structuring the output as per our desire whether it is a list, string, boolean or number it helps to get a structured output to use. With that I used OpenAI's gpt 3.5 turbo model to analyze the question papers and check it's factuality, integrity and at the same time match it with the paper that the professor uploads. problem it is solving: normally it takes weeks for teaching assistants or professors to grade a test paper and by the time we get the result and feedback we already have another test or quiz on our head so this helps to give feedback and result within minutes using AI, it will cut the cost of having to pay hefty amount of money too to the TAs and get the work done sooner so that students can work on it and improve on time.

Tutorial for Use and Best Practices:

I used MarvinAI's documentation i.e. https://www.askmarvin.ai/welcome/what_is_marvin/ and openAI's official documentation i.e. https://platform.openai.com/docs/introduction

Impacts on Learning:

It's gonna make a huge impact on students by providing them with feedbacks on the same day so that they can work on it and improve for the next one and at the same time save some funds of the college

Limitations and Ethical Considerations:

there are some limitations i.e. identifying the student's handwriting which is something even humans can't do sometimes hehe, but we have a solution for it that I might work on in future i.e. fine tuning an image processing large language model with bad handwritings Even validating the answers by AI was tough but that's when I thought that after AI checks it by it's own it can compare it with factuality with the other pdf with correct answer that the professor uploads

Link to Video Tutorial (Optional):

https://www.youtube.com/watch?v=OFbNeIdl7vE&t=5s

Inspiration

I get really tense after every quiz or mid term exams until I get my grades in hand but that takes weeks because there are 100s of students and the teaching assistants can't grade them all within days so that's when I got this idea to create a SAAS platform to help them overcome this problem and make it a lot more time efficient using AI

What it does

MarkSwift helps professors to grade all the student's exam papers using openAI and Marvin, we have to features

first feature is where professors can upload the sheet with the correct question answers and upload the file of the students and AI will go through it and give an output of the incorrect question answers with the expected ones and the marks that are deducted for it.
second feature is where the professor can talk to the document in order to check it's factuality or just grade it there too with custom answers from the UI itself, professor's can choose if they want the answer is string i.e. text or in boolean structure. ## How we built it
I used openAi's api with marvin to give the output a proper structure and ability to through the documents.
I used streamlit for seamlessly creating the UI and integrating it with the openAi gpt 3.5 turbo model ## Challenges we ran into
Streamlit was all new for me so I was getting this issue that if I click on the other input the previous input text was disappearing but due to the awesome streamlit community I figured out that streamlit basically re does everything whenever there is a change so in order to keep that input we should use the session_state from streamlit and that helped me to fix that issue.
Creating multi page web app was a bit challenging but with the help of some tutorials I was able to fix that too. ## Accomplishments that we're proud of I'm proud that I was able to make this project within 24hrs all by myself and made it work too at the same time as I expected it to be and the fact that I am solving an issue that I was facing makes me feel even more happy about it. ## What we learned
I learned how we can tinker around with AI to get structured outputs which are really necessary in order to save tokens and at the same time give straight forward answers.
I learned how to use streamlit from scratch. ## What's next for MarkSwift Next I am planning to make it to be able to write the answers into a excel sheet with the student's information and then the professors can download it, will be talking to my professor this week to see if he has any feedback on it or if he would like to be one of my beta tester.