During Online learning, there are several student's and teacher's problems:
- Lack of Interaction in Online Learning 85% students agreed. 71% find harder to understand the material in online learning.
- Lack of Reading Comprehension Skill in my country (Indonesia) Indonesians are still low in their reading ability.
- Heavy Duty of Teachers Teachers occupied with teaching, engaging with students, and administrative tasks. In Brazil 67% were anxious, 38% felt tired, and less than 10% were happy (WorldBank).
What it does
It have goal of helping students around the world (especially Indonesian students) to study during online learning through interactive learning environment. Right now it support 2 languages, English and Indonesian. It have motto "Sharpen your understanding with the help of AI". It has 4 feature those are:
- Q Gen (Main Feature) Generate list of question and answer from given text. This will help teachers create pop quiz from that text.
- Q Gen FlashCard (Main Feature) Generate questions and transform them to a deck of flashcard. This will help student rememorize their learning.
- AnswerIt Answer question based on given text. This will help student to get answer fast from given context such as text.
- SumUp Create summarization from long text.
How we built it
We use transformers with help of huggingface library to build model for generate question. We train the model using graviton EC2 instance (g4dn). We wrap that model with HTTP API using FastAPI. For the frontend, we use Next.js.
Challenges we ran into
- Training the model in ec2 instances. We have some problems to enable GPU for cuda.
- Running the model in ec2 using docker and FastAPI.
- Managing HTTPS connection. Enable HTTPS connection in docker environment with help Nginx
Accomplishments that we're proud of
- The model have high accuracy.
- The model have huge impact for students in learning.
- Success deploying machine learning model in EC2 instances.
What we learned
- We learned new EC2 instances group that is g4dn (high performance GPU and CPU core with lower cost) and use its architecture.
- We learned training and deploying machine learning models in AWS.
What's next for SQNA
- We would like to build more high availibilty and scalability for model endpoint in order to can be used by every students in Indonesia hopefully other countries freely.
- We would like to provide on-demand API for generating question.
- Gather more data for improving machine learning models and optimize model for reduce inference time.
- Improve web design for better user experience and add scoring system.