I’m making a ReactJS app for a project that users can type in their essay/piece of writing and when they click a button it’ll show a bunch of AI analysis. This analysis includes spellcheck, an estimated teacher score, and a chatbot area where users can ask questions on their essay and get feedback on it. I’ve made the spell check and teacher score area by using a Flask python server and some simple machine learning based on an essay dataset I found.

I’m a beginner in NLP and I’m using this project to learn it so I don’t know what I’m doing? Haha I’m just stuck finding a resource that can help me understand how to achieve what I’m doing that doesn’t say the same thing as the last 10 ones.

I’ve tried using my score to generate input like (if score < 70 then recommend better grammar) but it seems too static. I also tried using different readability score measures like flesh-Kincaid but the response it gave wasn’t as personalized as I want it to be. I want it to be generated uniquely based on the user’s essay.

I also tried using BERT (a language model- I think) but I’m confused on the data I need to finetune it. All the examples I’ve seen online are too specific and don’t relate to my problem, so I’m just lost finding out how to start solving the problem. Should I stick to static text generation or try with BERT? I’m worried I don’t have enough data to train it properly. Would it even help with the chatbot?