Automated SRAI embeddings (worker API)

Geobase can run the same OSM → H3 → join → CountEmbedder pipeline described in SRAI basics (Part 1) as a background job on the project worker, then insert vectors into geoembeddings via the same catalogue path as raster embeddings.

When to use this

You want hosted runs (no local Python notebook) with job status in Studio.
You accept live Overpass / OSM fetches (keep AOIs small; respect provider rate limits).

Worker task

Task name: createSraiEmbeddings
HTTP (OpenAPI): POST /jobs/createSraiEmbeddings on the project worker (same host pattern as createGeoEmbeddings).
MVP embedder: embedderProfile: "count" (CountEmbedder + OSMOnlineLoader), model catalogue name srai-count-osm-v1.

Request body (conceptual)

Field	Required	Notes
`tableName`	yes	Valid PostgreSQL identifier; destination table in `geoembeddings`.
`h3Resolution`	yes	Integer 0–15 (see H3 docs; tutorials often use 9–11).
`placeQuery` or `aoiGeoJson`	one required	Place string is geocoded via srai; or supply GeoJSON Polygon / Feature / FeatureCollection in WGS84.
`osmTags`	no	Tag filter, e.g. `{ "amenity": "bicycle_parking" }`.
`osmPreset`	no	`bicycle_parking` maps to the tutorial default.
`uploadArtifactsToStorage`	no	Same behaviour as raster jobs: optional GeoParquet upload after insert.

Studio

In Create GeoEmbeddings, choose OSM (SRAI) to submit the same job through the Studio API proxy (authenticated with the project service role).

Manual notebooks: Part 1, Part 2 (Hex2Vec is not automated in this MVP).
Worker implementation: apps/worker/tasks/sraiEmbeddings.ts, apps/worker/bin/create_srai_embeddings.py.

Automated SRAI embeddings (worker API)

When to use this

Worker task

Request body (conceptual)

Studio

Related