Late response, but maybe you'll still read it... We had some issues because there was no real security on the API of our NLP tooling. So we put NGINX in front

Late response, but maybe you'll still read it... We had some issues because there was no real security on the API of our NLP tooling. So we put NGINX in front of it to create a from of API key auth. NGINX would deny the larger messages by default, so we had to increase some parameter so it would pass the large documents. And indeed, since this was only relevant in the backend, it didn't matter. For the front-end we built our own API that was a lot more sensible.