map-search

Intent-aware local map search engine over OSM/Orbis NexVentura PBF data. Built with Python · Flask · SQLite FTS5 · Leaflet.js

---

Features

Feature	Details
Full-text search	SQLite FTS5 with unicode61 tokeniser + diacritic removal
Two ranking modes	V1: BM25 + popularity · V2: multi-signal (name match, place type, category, confidence, distance)
Tier-1 intent	Transport-hub context — "parking near Brussels Airport", "car rental nearest to TomTom Gent"
Tier-2 intent	General POI categories — "pharmacy near Gent", "restaurant in Brugge" (EN / NL / FR)
Reservation check	Async check per restaurant result: detects TheFork, OpenTable, Resengo, Zenchef, and 16 other booking platforms; clickable badge links directly to the reservation page
Reverse geocode	/api/reverse?lat=…&lon=… — nearest features to a point
User location	⊙ button uses browser Geolocation API for distance-aware results

---

Prerequisites

Python 3.9+
pip install flask osmium

An Orbis NexVentura PBF file (or any OSM PBF).

---

Quick start

1. Build the index

python3 build_index.py /path/to/region.osm.pbf search.db

For the Belgian extract (~110 M nodes) this takes ≈13 minutes and produces a ~1.6 GB search.db. The script skips nodes with no name / address / place tag, resulting in ~6 M indexed features.

2. Start the server

python3 app.py
# → http://localhost:5017

---

API reference

GET /api/search

Parameter	Type	Default	Description
q	string	—	Search query
limit	int	10	Max results (1–50)
type	string	—	Filter: address | poi | place | named
category	string	—	Filter by OSM category (e.g. restaurant)
lat	float	—	User latitude (enables distance scoring)
lon	float	—	User longitude
ranking	string	v1	Scoring mode: v1 or v2

Response

{
  "query": "brussels airport",
  "count": 10,
  "ranking": "v2",
  "intent": {
    "type": "park",
    "label": "Parking near",
    "icon": "🅿️",
    "anchor": { "name": "Brussels Airport", "lat": 50.8978, "lon": 4.4831, "id": "…" },
    "note": null,
    "suggestions": [{ "label": "🏨 Hotels", "q": "hotel near brussels airport" }]
  },
  "results": [ { "id": "…", "name": "…", "lat": …, "lon": …, "score": …, … } ]
}

GET /api/reverse

Parameter	Type	Description
lat	float	Latitude
lon	float	Longitude
limit	int	Max results (1–20, default 5)

GET /api/check_reservation

Parameter	Type	Description
url	string	Restaurant website URL

Response

{
  "reservable": true,
  "platform": "TheFork",
  "source": "url",
  "reservation_url": "https://www.thefork.be/restaurant/…"
}

source is one of url (the site itself is a platform page), link (platform link found in page HTML), or keyword (reservation keyword found in page HTML).

---

Index schema

CREATE TABLE features (
    id           TEXT PRIMARY KEY,   -- gers_identifier or "n/<osm_id>"
    feature_type TEXT,               -- address | poi | place | named
    name         TEXT,
    housenumber  TEXT,
    street       TEXT,
    city         TEXT,
    postcode     TEXT,
    category     TEXT,               -- OSM amenity/shop/tourism/… value
    place_type   TEXT,               -- OSM place value (city/town/village/…)
    phone        TEXT,
    website      TEXT,
    lat          REAL,
    lon          REAL,
    confidence   REAL,               -- Orbis confidence:feature (0–1)
    popularity   REAL,               -- Orbis popularity score
    search_text  TEXT                -- concatenated searchable text
);

CREATE VIRTUAL TABLE search_idx USING fts5(
    search_text,
    tokenize = "unicode61 remove_diacritics 1",
    content  = features,
    content_rowid = rowid
);

---

Kubernetes deployment (AKS)

The app runs on Azure Kubernetes Service. search.db is stored in Azure Blob Storage and downloaded by an init container at pod startup (~30 s for 1.5 GB).

Prerequisites

• AKS cluster access: az aks get-credentials --resource-group map-experts-tools-rg --name map-experts-tools --subscription orbis-poi-dev-1
• ACR: searchpoi.azurecr.io
• Storage account: poitfstate (blob container map-search-data, file search.db)

First-time setup

# 1. Upload search.db to blob storage
STORAGE_KEY=$(az storage account keys list --account-name poitfstate \
  --subscription orbis-poi-dev-1 --query "[0].value" -o tsv)
az storage blob upload --account-name poitfstate --account-key "$STORAGE_KEY" \
  --container-name map-search-data --name search.db \
  --file search.db --overwrite

# 2. Create Kubernetes secrets
kubectl create secret generic azure-files-secret \
  --from-literal=azurestorageaccountname=poitfstate \
  --from-literal=azurestorageaccountkey="$STORAGE_KEY"

SAS=$(az storage blob generate-sas --account-name poitfstate \
  --account-key "$STORAGE_KEY" --container-name map-search-data \
  --name search.db --permissions r --expiry 2027-01-01 --https-only --output tsv)
kubectl create secret generic map-search-config \
  --from-literal=DB_BLOB_URL="https://poitfstate.blob.core.windows.net/map-search-data/search.db?${SAS}"

# 3. Build image and deploy
az acr build --registry searchpoi --subscription orbis-poi-dev-1 \
  --image map-search:latest --file Dockerfile .
kubectl apply -f k8s/deployment.yaml -f k8s/service.yaml

Update workflow

# Local dev → test at http://localhost:5017
lsof -ti :5017 | xargs kill -9
python3 app.py > /tmp/map-search.log 2>&1 &

# Push to cluster when ready
az acr build --registry searchpoi --subscription orbis-poi-dev-1 \
  --image map-search:latest --file Dockerfile .
kubectl rollout restart deployment/map-search
kubectl rollout status deployment/map-search

Live URL: http://4.207.13.41

Note: Secrets (azure-files-secret, map-search-config) must be created manually in the cluster — they are never committed to this repository. See k8s/pvc.yaml for the exact commands.

Project structure

map-search/
├── app.py              Flask API + intent engine + reservation checker
├── build_index.py      PBF → SQLite FTS5 index builder
├── Dockerfile          Container image definition (search.db excluded)
├── k8s/
│   ├── deployment.yaml Kubernetes Deployment with init container (blob download)
│   ├── service.yaml    LoadBalancer Service (port 80 → 5017)
│   └── pvc.yaml        Reference only — secrets creation commands
├── templates/
│   └── index.html      Leaflet UI (map + sidebar + intent banner)
└── search.db           Built index (git-ignored, ~1.6 GB for Belgium)

---

Supported booking platforms

TheFork · LaFourchette · OpenTable · Quandoo · Bookatable · Zenchef · Resengo · Formitable · CoverManager · Resy · SevenRooms · Eat.be · Resto.be · Bookingkit · Planity