As mentioned in connector document to specific Schema of dataframe and avoid inference, We have to provide case class. But it some usage cases (same as mine) it not feasible and I prefer to infer schema with high rate of samples for the first time, then cache it and use cached schema.

For that I should use following way in my source code which is not clear enough and some how it should be away from my eyes.

builder().sparkSession(session).readConfig(readConfig).build().toDF(schema) 

I would like to have it simple as possible, somehow like 

session.loadFromMongoDB(readConfig, schmea) 

 If you think it's good way, I can create pull request for it.