The reasons why one would choose to use RNI as standalone software vs a plug-in should be founded in the desired features and your organization's product knowledge. Standalone RNI favors indexing only names, whereas the plug-in is more appropriate if other fields (e.g. address) must be associated with each name.
Standalone RNI uses Lucene directly. The RNI plug-ins are made specifically for a search product, e.g. Solr, Elastic.
In all cases, there are two steps:
- High-recall search
- Re-ranking
In the case of standalone RNI, the High-recall search uses Lucene's native search function to return all names that might possibly match the query. Since Lucene name matching is not nearly as accurate as RNI, the bar for these matches must be set very low. Additionally, extra search fields (e.g. Ascii Folding) are used to catch even more results.
In the case of the RNI plug-in, the High-recall search is performed using the chosen search product (which typically also uses Lucene).
The Re-ranking step is the same in standalone RNI as in the plug-ins. In both cases it takes each search result from the first step and determines the RNI pairwise-match (a value from 0 to 1), and returns the results in the new order.