Show HN: Fast deterministic natural language search for OpenStreetMap data
I want to show you HippoMaps. It is a fully featured map and navigation app built on OpenStreetMap with a custom Overpass API engine for really fast search of POIs, areas and ways.My aim with HippoMaps is to get to a level where I don't need Google Maps anymore. The biggest hurdle IMHO with OSM based apps is the lack of nice and fast search (where GMaps excels at). To get there I started working on a deterministic (as in no LLMs used) parser of queries that translates them to the Overpass QL.Search queries are split into tokens and the algorithm tries to guess "what" you want to search and "where". The former comes from a word list (known brands, sports, OSM tags) and the latter from the geocoder service (Photon).You can get quite creative like "cafes in Zurich by the Limmat", "peaks over 4000 m in Valais", "pharmacies open now in Bern". It's good for discovering hiking trails, e.g. "hiking trails from Arth-Goldau to Rigi Kulm". It knows how to find stuff between two places, next to a place, near a place and so on. You can stack those as well like "bridges over the danube in serbia 50 km from novi sad".Except the search it has almost all of the features you would expect from a map like turn-by-turn navigation (Valhalla) with voice (Piper), public transport navigation in Europe (MOTIS), live traffic (only a few countries so far).The app is GPL-3.0: https://github.com/cdutsov/hippomapsI'm looking forward to feedback on searches that work or don't work for you + overall comments.
113 points by - 32 comments
The README suggests that the app is published to F-Droid:
> F-Droid builds the app from the tag v<versionName> (recipe metadata/ch.hippomaps.app.yml in fdroiddata) and publishes it with our signature: the build is reproducible, so F-Droid copies the signature of our APK onto its own build and publishes only when the two are the same.
but I see no such app: https://search.f-droid.org/?q=hippo&lang=en
nor do I see an issue requesting its inclusion: https://gitlab.com/fdroid/rfp/-/work_items?state=all&search=...
(I also didn't see it in the Google Play Store, though I wouldn't be installing it from there anyway.) Is there another place I should be looking for the app?
"A cafe in {my town} with a nice view". Returned 16 hits, but not ones with a nice view. I guess there's no way for it know. Unless you can get access to online reviews.
Edit: It chose a "node" of a "viewpoint" and found cafes within 1.1km of that. Interesting, but not correct.
wdym?
I immediately noticed some quirks in the parsing - for example: https://hippomaps.ch/?q=campsite+within+100+ft+of+a+mine+in+...
In that query, it should interpret "campsite" as (tourism=camp_pitch || tourism=camp_site), not camp_site alone, and "mine" as (industrial=mine || historic=mine), not minefield! (The camp_pitch vs camp_site distinction is very annoying and I wish those had better names.)
I love the features mappings you do, then you wrapped a nice UX around it. I did try to find dispensaries in Stamford, which here means cannabis retailers. It mapped that to "pharmacies" instead, so I got a CVS drug store. It got my Stamford (CT) correct and suggested a few other Stamfords (like VT).
Since it is Golang, so you might be interested in our ntmaps library. It is MIT-licensed, so should be compatible. It would let you make terminal apps with this (or browser, but you handled that already).
https://nimbleterminal.github.io/ntmaps/?demo=explorer
"sparsely populated urban residential areas with >75% greenery nearby".
I was able to make it work.. tested it on a small area of 50x50 km2.
But the costs were insane.. first to populate the database, and then to host the query engine.
"Houses more than 300 m away from pubs and more than 1 km away from train stations"
This already works
The parser restated it clearly, but returned just one result: Felix at Bellevueplatz 5.
Nearby, the map shows several other café-like places, so I wasn’t sure whether the search was intentionally strict or only matching a narrow OSM category.