Open source · MIT · Built on OpenStreetMap

Address search,
streamlined for Pakistan

A streamlined, open-source POI search for Bykea-style addresses. We pull raw data from OpenStreetMap, clean it with Gemini, and serve fuzzy, multilingual results — so riders find places, not friction.

گلبرگ liberty
  • Liberty Market, Gulberg IIILahore · 1.2 km
  • Liberty Roundabout (لبرٹی چوک)Lahore · 1.4 km
  • Hussain Chowk, GulbergLahore · 2.1 km
What it does

Search that understands the way people actually type

OSM as the source

Raw POIs pulled live from OpenStreetMap's Overpass API — community-owned, always growing.

Gemini-cleaned data

We use Gemini to normalize names, fill gaps, transliterate, and dedupe noisy entries.

Fuzzy & fast

Typo-tolerant matching with token-set + Levenshtein ranking. Sub-100ms responses.

Urdu + Roman + English

Mix scripts in one query. 'گلبرگ liberty' just works.

Distance-aware

Ranks results by proximity to the rider, not just text similarity.

Drop-in REST API

Simple endpoints, JSON in, JSON out. Self-host or use our hosted instance.

Under the hood

From raw OSM to searchable POIs

01

Pull from OSM

Query Overpass API for POIs in target regions — markets, landmarks, schools, fuel stations.

02

Clean with Gemini

Normalize names, transliterate Urdu ↔ Roman, fill missing categories, cluster duplicates.

03

Serve fuzzy results

Index into a typo-tolerant search engine, ranked by relevance + distance.

The problem

Finding Bykea addresses shouldn't be this hard

  • Raw OSM data is noisy, inconsistent and multilingual
  • Users type messy queries — short, mixed-script, abbreviated
  • Existing geocoders miss Pakistani POIs and informal landmarks
  • Riders waste minutes searching — that's lost revenue
"g9 markaz"
→ no results
"lhr airport t1"
→ wrong city
"گلبرگ مین بلیوارڈ"
→ unmatched
"DHA phse 5"
→ 0 hits