It uses extensive NLP and ML techniques at its backend for answering descriptive type queries ( currently, the model has been trained only for Data science and Machine Learning Domain ).
Your project is really interesting and I am wondering what the connection is between the classification and doc2vec top 5 document retrial.
It looks like two independents tasks for me. I am guessing are you planning to use the classification to shrink the document size before you retrieve the top 5 documents?