@rajnish0395 Just to add further ..

Chatbots come in 2 flavours.
i) generative
II) retrieval - This is what most use (e.g. wit.ai) and are widely successful

Seq2Seq is ideal for generative models but very difficult to make through.
You could try a retrieval model.

In retrieval model, you would have a few pre-stored responses (these could be something that user has responded in the past). You would use a RNN to find a matching response.

Have a look at quora's duplicate q finding mechansism. Your model would work same way but once you find a similar question or input text, you would respond with an intent.
[https://engineering.quora.com/Semantic-Question-Matching-with-Deep-Learning]

Again this is just one example of retrieval model. There could be other methods. You could also search in Google scholar for latest techniques and then search for corresponding code in github.