It sounds like a small milestone, but there was a ton of work to do this. Initially, I used ready-made packages to handle text classification. They were ok, but they also classified irrelevant data.
I decided to build my own neural net. I didn't know ML. I went with TensorFlow.js. Tried to build one for 3 months. Some of them were ok, but also missed a lot. Then, after learnings, I went with a pre-trained model.
Eventually, I'm using it.
Other updates: