📋 More geo with bounding

Below is a MongoDB Atlas Search index and $search query sample for doing vector search inside the $search stage with a geo bounding filter.

Important: because you want a geo filter such as geoWithin, use Atlas Search $search.vectorSearch, not the standalone $vectorSearch aggregation stage.

---

1. Atlas Search Index

Example index name: places_vector_geo_idx

json

{

"mappings": {

"dynamic": false,

"fields": {

"embedding": {

"type": "vector",

"numDimensions": 1536,

"similarity": "cosine"

},

"location": {

"type": "geo"

},

"title": {

"type": "string"

},

"description": {

"type": "string"

}

}

}

}

Assumed document shape:

js

{

_id: ObjectId("..."),

title: "Some Place",

description: "A description of the place",

embedding: [/ 1536-dimension vector /],

location: {

type: "Point",

coordinates: [-73.9857, 40.7484] // [longitude, latitude]

}

}

MongoDB GeoJSON coordinates are always:

js

[longitude, latitude]

---

2. Vector Search with Geo Bounding Box Filter

Use this when the user provides a rectangular area, usually as:

• southwest / bottom-left point
• northeast / top-right point

js

const queryVector = [

/ embedding generated from the user's query text /

];

const bottomLeftLongitude = -74.10;
const bottomLeftLatitude = 40.60;

const topRightLongitude = -73.70;
const topRightLatitude = 40.90;

db.places.aggregate([
{
$search: {
index: "places_vector_geo_idx",
vectorSearch: {
path: "embedding",
queryVector: queryVector,
numCandidates: 200,
limit: 10,
filter: {
geoWithin: {
path: "location",
box: {
bottomLeft: {
type: "Point",
coordinates: [
bottomLeftLongitude,
bottomLeftLatitude
]
},
topRight: {
type: "Point",
coordinates: [
topRightLongitude,
topRightLatitude
]
}
}
}
}
}
}
},
{
$project: {
title: 1,
description: 1,
location: 1,
score: {
$meta: "searchScore"
}
}
}
]);

---

3. Same Query with Placeholder Coordinates

If the user will provide exact lat/long values, structure it like this:

js

const queryVector = [/ generated query embedding /];

// User-provided bounding area
const southwest = {
lat: USER_PROVIDED_SW_LAT,
lng: USER_PROVIDED_SW_LNG
};

const northeast = {
lat: USER_PROVIDED_NE_LAT,
lng: USER_PROVIDED_NE_LNG
};

db.places.aggregate([
{
$search: {
index: "places_vector_geo_idx",
vectorSearch: {
path: "embedding",
queryVector: queryVector,
numCandidates: 200,
limit: 10,
filter: {
geoWithin: {
path: "location",
box: {
bottomLeft: {
type: "Point",
coordinates: [southwest.lng, southwest.lat]
},
topRight: {
type: "Point",
coordinates: [northeast.lng, northeast.lat]
}
}
}
}
}
}
},
{
$project: {
title: 1,
location: 1,
score: {
$meta: "searchScore"
}
}
}
]);

---

4. Alternative: Circle Around a Lat/Long

If the user gives a center point and radius instead of a bounding box:

js

const queryVector = [/ generated query embedding /];

const centerLatitude = 40.7484;
const centerLongitude = -73.9857;
const radiusMeters = 5000;

db.places.aggregate([
{
$search: {
index: "places_vector_geo_idx",
vectorSearch: {
path: "embedding",
queryVector: queryVector,
numCandidates: 200,
limit: 10,
filter: {
geoWithin: {
path: "location",
circle: {
center: {
type: "Point",
coordinates: [centerLongitude, centerLatitude]
},
radius: radiusMeters
}
}
}
}
}
},
{
$project: {
title: 1,
location: 1,
score: {
$meta: "searchScore"
}
}
}
]);

---

Key Notes

• Use an Atlas Search index, not a Vector Search index.
• The vector field uses Search mapping type "vector".
• The geo field uses Search mapping type "geo".
• GeoJSON coordinate order is always:

js

[longitude, latitude]

• The result score should be projected with:

js

{ $meta: "searchScore" }

not:

js

{ $meta: "vectorSearchScore" }

• numCandidates should generally start around 20 * limit, then tune for recall and latency.