Hello Raj,
Being a Machine Learning enthusiast, just now I've completed developing a PoC for a text-based Search Engine to identify documents from the Document Management System which matches a given search sentence. As the documents in the DMS were mostly word files with somewhat similar structure, I first read the files using an existing Python library, then perform a bit of preprocessing to identify relevant text from the documents and then the Universal Sentence Encoder to perform the matching process. The core of the Search Engine was developed using Tensorflow Hub using Python 3.7
As it seems that, you want to develop a metasearch engine, I would like to go through further details of the project and see if I can help you with the same.
Please reply to this proposal, if you find it useful.
Regards,
Atul Rajdhar