Prova

Inspiration

College applications are already very stressful for high school students. Ontop of many essays, most colleges require or prefer standardized test scores to be reported. SAT Subject Tests and AP Tests are notoriously difficult to prep for and data and materials are limited. Avi had the idea to use preexisting data (from the Summit Learning Platform) to make significantly more accurate and much more time-friendly test (compared to Practice Exams, which are already hard to obtain) to predict how well one will score.

What it does

One aspect (the least significant) is a web crawler made specifically to extract data from the Summit Learning Platform. It is likely the least significant because we want to expand to multiple different online school grades systems. We collect hundreds of thousands of data points from the crawler, which are fed into a sequential neural network (after using preprocessing best practices) that outputs the scores for each individual test being analyzed. The data points used include scores on online lessons, standardized rubric-based project scores, and even how early students are completing lessons on the PLP. These let the network use categories to further the algorithm. It predicts the scores and then suggests which ones are recommended to take. (We also have an autofilling search to make it easy for the user to see if and why the AI is not recommending a certain test. They can do this if they feel confident in their work but it hasn't shown in their grades yet. An additional factor in adjusting the predictions is a short diagnostic given to the user after they click on a test. We have not written the diagnostic questions due to time constraints and it wouldn't help I showing the technology. Resources are linked, but we were also in the process of adding a teacher login so that teachers authorized by an authorized superuser can add their own resources that they have created specifically for the tests. There are also slight recommendations saying if a given score is strong or not.

How we built it

We built using HTML/CSS, JS, Node JS, and Tensorflow.

Challenges we ran into

fixing asyncronity issues with training the model in tfjs

Accomplishments that we're proud of

Neeraj Rattehalli is proud of creating his first website using several languages. Avi is proud of doing so much work on the full stack in a limited amount of time with nearly no preparation. Avi is also proud of the small features added to enhance the user experience.

What we learned

Neeraj learned about every aspect of web development. Avi learned how to clean up a UI not just for aesthetics (which can be done by a lot of people) but for user efficiency, which is much more critical.

What's next for Prova

We plan to make this full-fledged with a live retraining and an incredible UI. We would also like to finish the teacher login and find other methods of collecting data, as crawlers are technically volatile.

wtsit.tech

Built With

• colab
• colaboratory
• css3
• html5
• javascript
• keras
• node.js
• python
• tensorflow
• tensorflowjs
• tfjs