Rework

Inspiration

The education system is broken and teachers are under-appreciated. So, we wanted to create something to help teachers. We spoke to a teacher who told us that a lot of her time is spent editing students’ work and reviewing tests. So, we started thinking about how we could help teachers save time, while building a straight forwards user friendly solution.

What it does

Rework enables teachers to produce easier and harder versions of the same test based on the test scores of past class sections. Teachers input a photo of the test with questions & answer key, the average student score for each question, and the desired class average for each question. Rework then looks at the test questions where the class average was far below or above the desired average, and makes the harder or easier based on how much above or below the average it was.

How we built it

We built the backend of our product using Python and Flask paired with a basic frontend built with HTML, CSS, and JavaScript. Our program utilizes an OpenAI API to access GPT-3 for the main functionality in generating normalized test questions. We also made use of flask and a virtual environment for our API along with leveraging some OCR software to read our PDF inputs. We built our project centered around the normalization of test questions and then added functionality from there namely interacting with the service through pdfs.

Challenges we ran into

Our team faced a multitude of setbacks and problems which we had to either solve or circumnavigate over the course of the hackathon. Primarily our product makes use of an API connected to GPT-3, working with this API and learning how to correctly prompt the chatbot to obtain desired responses was a challenge. Additionally, correctly fragmenting our project into manageable goals proved to be important to time management.

Accomplishments that we're proud of

We created an MVP version of our product where a .json file would be needed to submit the test questions and answers. We wanted to finish this quickly so that then we could use the rest of our time implementing an OCR so that teachers could simply submit a picture of the test and answers and the questions would be read and parsed into a readable format for Rework to be able to understand, making the life of the teacher significantly easier. We are proud that we were able to add this extra OCR component without having any previous experience with this.

What we learned

Our group had a wide range of technical abilities and we had to learn quickly how to use all of our strengths to benefit our group. We were a fairly new team, the majority of whom were at their first major hackathon, so there were lots of growing pains. Having each team member understand the technologies used was a important task for Friday, as well as organizing ourselves into roles where we could each excel with our diversity of experiences and comfortable languages. Almost all of us had little expertise with front-end development, so that is the technical area where we improved most—along with creating a full project from scratch without a framework.

What's next for Rework

In the future with more time we would like to expand on the feature offering of Rework. Namely, the inclusion of automatic grading software after we convert the image would allow for a more wholistic experience for the teachers, limiting their number of inputs while simultaneously increasing the functionality. We would also like to implement a more powerful OCR such as mathpix, ideally one that is capable of latex integration and improved handwriting recognition, as this would give more options and allow for a higher level of problems to be accurately solved. Ultimately, the ideal goal for the program would replace Chat-GPT with lengchain and GPT-3, as this allows for more specialized queries specifically for math enabling more accurate responses.