What would you like to see in the FIAS service?

Question

C++ / C#

I am slowly sawing the FIAS service, not official, but my own. What would you like to see in it. Implemented on net core, it will work on Debian and windows accordingly. Supported databases Postgres MySQL MS SQL. At the moment, I reversed their schemes to save space, implemented a loader, pieces of old code for rest requests that display the country, region, city.

What I want to implement is the omission of unnecessary data, such as regulatory documents, apartments, and other stuff.

In general, share your thoughts on what you need.

UPD.
The main repository, which will be supplemented in the future
https://github.com/vkorotenko/FiasServer

Associated repository, offline address parser, size on disk ~ 180 mb. Size in memory ~ 600 MB.
https://github.com/vkorotenko/NAddressParser

Supported frameworks

• netstandard1.3
  
• netstandard2.1
  
• net45
  
• net46
  
• net472
  
• netcoreapp1
  
• netcoreapp2
  
• netcoreapp3
  
• netcoreapp3.1
  

In general, we are moving slowly, In the first, everything is close to the stubs.
Implemented file uploader with versions.
Parser and loader to the database, so far only dictionaries

In the second, all the logic is up to the streets, unfortunately the current project contains newer addresses, so I switched to it.

UPD. Well, the import is complete.
Now the plans include adding a plugin for MySQL and a muzzle for queries.
The stack is something like this .NET Core REST api, swagger as interfaces documentation.

For the first time, a static list of users and admins, vuejs + bulma as an interface

UPD: 3
So for today, a client has been added to vue Identityserver4 with a few stubs. In general, I'm starting to collect everything for sale.
Who needs regions and cities, this is already present.

Answer 1

See what dadata provides . They have something that can be repeated in their service, for example, address normalization.
Another cool thing for all kinds of voice assistants is correlation sampling.
Let me explain.
The user is talking to the bot. It refers to Paris or the village of Olkhovatka. There are more than one such settlements. Especially Olkhovatok. However, some alder trees from the user are thousands of kilometers away, and one is ten kilometers away. Which one is the user most likely referring to?
So, we have several reference addresses: the home address of the user from the profile, the addresses of the last weather requests from the user, and so on.
Now the user asks where is the nearest five in the Olkhovatka area.
We look for it in the database and, according to the correlation base, find the most relevant in order of decreasing relevance. We assume the most probable and, if the second and third relevance index is too close to the first, then we specify: "is it in such and such an area which?", the user says, not in such and such another.
Such and such a district, as a separate address, gets into the user's POI cache, and after the next request of Olkhovatka in the address database, we line them up with a new relevance index. If again the rejected option is in the first place, then we skip it and make the following clarification.
I wrote chaotically, but this is not enough.
In general, it would be cool to have a base of aliases and toponyms synonyms.
Peter, St. Petersburg, Leningrad - these are aliases of one locality.

Answer 2

In the furnace network API - make a standalone solution on sqlite.