We're building a chatbot to help people learn english. It runs on chat platforms like Facebook messenger, wechat, telegram.
Currently we have live teachers chatting with students.
I want to automate the process of matching user comments to suitable replies.
We have a "training set" of previous conversations.
We want to use this to try and automate the conversations with the teachers.
The first layer could perhaps just "suggest" replies, next layer would be to send the reply right away if we have a high enough confidence level.
I understand we could do this with a classifier, or a simple neural network.
I'm looking for advice on designing a system that would enable this, and for someone to implement a first working build.
If suitable, we could use the "NLP as a service" cloud products out there, or else an open source library running internally.
Please write back with info on similar projects you've built, a general outline on what tech you would use to solve the problem, and any questions you would need me to answer to be able to estimate a time and cost for developing this feature.
I would like to develop an initial take on a fixed budget, but then with the right person this would be on-going work to improve the features and conversational aspect of the system over time.