GeoIntel release gates / Compile, test, contracts and builds (push) Successful in 1m49s
GeoIntel release gates / Python and npm vulnerability policy (push) Successful in 21s
GeoIntel release gates / Production AI image, SBOM and container scan (push) Successful in 5m39s
GeoIntel release gates / Deploy exact gated revision to Unraid (push) Failing after 58m43s
1.9 KiB
1.9 KiB
GeoIntel Kempen — M2 Engineering Package
This milestone turns the M1 Codex-ready blueprint into an engineering-grade repository specification. M2 adds decision records, explicit domain boundaries, event contracts, model registry rules, API response contracts, test matrices, and release conventions.
Purpose
Codex should not invent architecture. It should implement the architecture defined in this repository.
M2 makes the following decisions explicit:
- FastAPI is the backend application framework.
- PostgreSQL + PostGIS is the source of truth for relational and spatial data.
- Redis + RQ is the V1 background job mechanism.
- Local filesystem storage is the V1 object storage layer, with a clear path to MinIO/S3 later.
- GRB is the primary authoritative reference source for Flemish building and base-map validation.
- OSM is useful as supplemental and fallback context, not as the primary QA reference when GRB is available.
- YOLO/Ultralytics is the V1 object detection runtime.
- SAM or YOLO-seg is the segmentation direction, but segmentation can be introduced after the detection foundation is stable.
- Reports, chat/copilot, LiDAR, MLOps, and QGIS plugins remain secondary to the GeoAI pipeline.
Milestone status
M0 froze the product direction. M1 made the repo Codex-ready. M2 makes the repo engineering-ready.
What Codex should build first
- Backend foundation.
- Database models and migrations.
- Dataset intake and metadata extraction.
- Area/project management.
- Raster/vector processing contracts.
- Object detection job skeleton.
- QA/QC calculation skeleton.
- Frontend workspace wired to real API contracts.
What Codex must not build yet
- Full LiDAR point cloud processing.
- Custom model training.
- Multi-user authentication.
- QGIS plugin.
- Production cloud deployment.
- Complex report designer.
- Large demo datasets committed into git.