Autocomplete Improvements v2

Track milestone here

Autocomplete has fallen behind search and needs some love. Here are the things we need to work on to get it inline with the other endpoints.



  • synonyms for the street layer


  • 'Union Square' POIs should sort before addresses on the street 'Union Square'
  • various issues related to sorting when focus.point API used

New features / Research

  • improve the parsing engine, attempt to reduce use of addressit

  • investigate the new ES2+ FST, the query language and potential performance gains.

    • what is the effect on build times?
    • how much flexibility do we have for analysis? Does this mandate a certain version of ES?
  • enable interpolation for autocomplete queries.

  • enable language translation features for autocomplete.

  • remove admin areas/ address records from elasticsearch - to reduce index size and increase performance.

    • requires some research in to how many street names are in OA but not OSM and a plan to manage this.
  • match admin fields by id instead of by string

  • consider that alt names for venues/streets will be coming the in future and try to account for that in the data model.


  • fix known bugs (where major refactoring is not necessary)
  • feature catchup (interpolation, language etc. where possible)
  • complete the schema changes for /search (filter admin via id)
  • upgrade elasticsearch

  • concurrently
    • research FST, create performance benchmarking tooling
    • research placeholder autocomplete