The vision · AI Slam

Every survey, searchable. Every defect, detected.

AI Slam is the platform that closes the loop on marine survey intelligence. Survey Ops ingests and catalogs every capture from field to archive. Crackz runs pixel-level defect detection on that catalog. Together they turn raw survey data into actionable knowledge.

The two halves

One platform. Two engines.

Each product ships independently today and is designed to interoperate as part of the AI Slam platform.

Survey Ops Portal — capture catalog and project map view
Data foundation · Survey Ops

Ingest. Catalog. Retrieve.

Survey Ops is a local-first data catalog and ingest platform for marine survey data. Every capture from field device or vessel travels through inbox staging, operator approval, and canonical project storage — fully catalogued, geotagged, searchable, and ready for downstream analysis.

  • · Resumable field upload over Starlink (tus protocol)
  • · Inbox + operator approval workflow
  • · USBL geotagging — writes real GPS positions onto images
  • · Geospatial map view (PostGIS + Leaflet)
  • · Natural-language RAG search with source citations
  • · Offline-first — runs on operator hardware, no cloud dependency
Crackz detection results — per-image severity grid with defect overlays
Detection engine · Crackz

Detect. Segment. Report.

Crackz applies deep-learning semantic segmentation to find cracks, corrosion, spalling, and weld defects at the pixel level — across concrete harbor piers, bridges, factory surfaces, and energy assets. Pre-trained models deploy in days; custom models reach 90%+ accuracy on your domain data.

  • · Pixel-level defect segmentation (cracks, corrosion, spalling)
  • · Pre-trained checkpoints or custom-trained models
  • · GeoJSON output ready for GIS pipelines
  • · Severity scoring — High · Med · Low · Clean
  • · Audit-grade Word/PDF reports
  • · 90%+ detection accuracy on validation datasets
The integration

From field to finding — one continuous flow.

This is the architecture we're building toward. Today each product ships independently. The designed seam between them is how AI Slam becomes the operational system for marine survey intelligence.

00 Field capture
01 Ingest & catalog
02 Geotagged frames
03 Crackz detection
04 Annotations back
05 Searchable knowledge

Field capture

The survey vessel or ROV captures images and video. Survey Ops accepts resumable uploads — over Starlink or on-site — so no capture is lost to a bad connection.

Ingest & catalog

Packages stage in an inbox. An operator approves and the AI pre-fills the metadata. Captures move into canonical storage: customer / project / site / date, immutable.

Geotagged frames

USBL track data writes real GPS positions onto images. PostGIS stores the geospatial record; Leaflet shows every capture on a map. Embeddings make the whole archive semantically searchable.

Crackz detection

Survey Ops hands geotagged frames and their catalog context to Crackz. The detection engine runs pixel-level segmentation — cracks, corrosion, spalling — on every image.

Annotations back

Detection results, GeoJSON boundaries, and severity scores flow back as structured catalog annotations, linked to the original captures and their map positions.

Searchable knowledge

The entire survey — raw captures, geolocated positions, detection findings, and reports — is queryable in natural language. 'Show all high-severity pier scans from 2024' becomes a real query.

Seeing it together

Catalog on the left. Detection on the right.

Survey Ops shows every capture on a geospatial map, with natural-language search over the archive. Crackz surfaces per-image defect findings with severity scoring. AI Slam is the bridge.

survey-ops · project detail
Survey Ops Portal — project detail page showing SS Franken survey with geospatial map and metadata
Survey Ops · catalog & geospatial map
crackz-gui · detection run
Crackz detection run — severity grid with annotated defect images
Crackz · defect detection results
Vision · in progress Platform status

An open architecture built to interoperate.

Both products ship today as independent systems. Survey Ops is in active production use for marine survey data management. Crackz is in production across harbor infrastructure, bridge inspection, and manufacturing QA. The platform integration — automatic handoff from catalog to detection engine and annotations back — is the designed next phase of AI Slam.