Hi @Rendiyono Wahyu Saputro, I answer point to point: 1. You can use whatever you prefer. I used NIFI in a similar project for get and model the data from twitter 2. If you use NiFi you can process the data (the twitter API return a JSON file) to set the attributes with the values of JSON node. I suggest you to index data to Solr and use Spark for querying to Solr. After you should process the data in Spark for find the user with more retwitter tweets, for example. 3. In this page we used Storm to querying and generate JSON, for minimize the traffic to the SolrCloud (We don't have more CPU), but if you want you can queryng directly solr and return a JSON file for your app. My idea is: a- I register to your apps b- I insert my twitter account c - the app send to NiFi my personal information to follow me (add my userName in the getTwitter-processor filter) d - NiFi send to Solr my tweets e - Spark querying Solr for my username and process my data f - Spark send the result to another Solr collection (coll_B) g - the application requires the data processed to coll_B of solr in JSON format All components that I mentioned run on HDP in cluster mode. The size of the cluster depend on different factors: - how many people will use your app (many requests to solr require a lot of CPU and RAM) for example for ingest (avg) 800 tweets we have 3 NiFi worker with 3GB of Xmx - how heavy data processing? - It must be in real time? These and more are all things to be assessed for deploy a cluster. ... View more