Varasi · وارسی

AI-powered geospatial Change Detection platform. A Raster Catalog + Analysis Engine — references imagery in external storage, does not own it.

---

What it is

Continuously monitor large areas from satellite / aerial / drone imagery, automatically detect changes, classify them, and alert users. Open-source, cloud-native, multi-tenant.

Architecture (short)

Varasi is modular services, not a monolith. It integrates the proven eoAPI cloud-native stack for catalog + tiling rather than reinventing GDAL/GeoServer:

• web — Next.js dashboard (MapLibre, dark mode, EN/FA RTL).
• control-plane (Go) — auth, RBAC, multi-tenancy, jobs, alerts, WebSocket, external REST API.
• catalog (eoAPI) — stac-fastapi (search) + titiler-pgstac (COG tiling) + tipg (vector tiles).
• ai-worker (Python) — Open-CD / TorchGeo change detection + classification.
• ingest-worker (Python) — metadata-first scan of providers → STAC Items (no pixel copy).
• stores — PostgreSQL + PostGIS + pgSTAC, Redis, MinIO/S3 (derived artifacts only).

Full design: PLAN.md · Roadmap: SPRINTS.md

Quickstart

cp .env.example .env
make up        # postgis+pgstac, redis, minio, stac, raster, vector
make health    # wait for healthy
make bucket    # create MinIO derived bucket
make seed      # ingest sample public COGs

Endpoints: STAC :8081 · Raster :8082 · Vector :8083 · MinIO console :9001.

Concepts

• Raster Profile = a STAC Item (footprint, CRS, bands, cloud%, sensor, datetime, asset URIs).
• Virtual Dataset = a STAC Collection (Sentinel, drone flights, yearly mosaic...).
• Provider = anything that can be turned into a STAC Item (filesystem, S3, HTTP, COG, STAC, GeoServer...).

Status

Early build. Following SPRINTS.md, vertical-slice first (ingest → catalog → map).