Hello, I started this project as part of a Hackathon (BuilderHacks S2) and would like sharing it with you.
The project
The basic idea is developing chatrooms that allow friends to study together by submitting questions with a ML model in the backend verifying the answers. The latter is important because oftenly, answers are checked like this:
NLP and other technologies like CAS are used to smartly grade submissions like so:
Grading questions
Hence, several techniques are used to verify answers:
- Text strict: As you might expect, verifies that expected answer and submission match exactly.
- Text medium: Transforms to lowercase and strips away punctuation (and parentheses, apostrophes, etc) to verify that expected answer and submission match loosely.
Text similarity: Uses NLP (Though spacy) to verify that the similarity score for answer and submission is above 0.8
Scalar: Uses quantulum3 to identify scalars with unit and number and verify the answer matches submission (e.g., 10 cm = 10.0 centimeters)
Math: Uses CAS to make sure mathematical expressions match. (e.g., "(x*x)/2" == "0.5*x^2")
Other features
Additionally other features include KaTeX support, a drawing canvas and chatroom generation.
In the future
I developed the project in under two days as part of the hackathon but I intend to take it even further in the future, solving many bugs (and there are many) as well as integrating new features. I am open to feedback as well as working together with other people to make it happen.
Technologies used
- Frontend: React
- Frontend (real time communication): hop.io
- Backend: Python
Please let me know if you have any feedback and if you'd be interested in participating.
Devpost submission with more info: https://devpost.com/software/chat-edgwsa?ref_content=my-projects-tab&ref_feature=my_projects
Top comments (1)
Hi there, this post might fit better as a DEV Listing. It’s a dedicated area of the platform where community members and organizations are encouraged to publish information related to events, products, services, job listings, and everything in between.