Dark vessel candidates
Radar-only contacts, AIS gaps, missing identity, and track continuity become review-ready evidence instead of scattered operator clues.
Hubble StackBrief the teamAMSAS / Passifera
Maritime AIfused atwatch speed.
Hubble Stack AMSAS is an AI-enabled Maritime Security Analytics System designed as a non-intrusive decision-support layer over the existing Coastal Surveillance Network. It turns fragmented sensor feeds into one risk-ranked operational picture.
Common Operational PictureVOI-09 / High
Unmatched contactAIS silent + radar continuity
Risk score92
Evidence links7
Latency classLive
18 mopilot-ready timeline
TRL 7-8field validation target
AI + MLfusion and analytics core
On-premsecure deployment posture
Inputs
One fused vessel truth from every available signal.
Radar
AIS
EO / IR
VHF
PANS
Weather
Registries
Watchlists
The mission
You cannot intercept what the system cannot correlate.
Spoofing and mismatch logic
Duplicate MMSI, invalid identity, EO class mismatch, route conflict, and registry inconsistencies are surfaced with confidence.
Suspicious behavior patterns
Loitering, geofence entry, route deviation, unexpected meetings, and abnormal movement are ranked as Vessels of Interest.
VOI ranked alert
HIGH / PASSIFERA-82
Explainable risk, not black-box alarm noise.
AMSAS combines deterministic rules, geospatial logic, statistical anomaly detection, and ML models. The system prioritises evidence and keeps authorised watchkeepers in control.
Radar track continued after AIS lossRestricted geofence entry confirmedPANS route mismatch foundVHF meeting keywords detected
Multi-sensor fusion
Associates radar, AIS, EO, VHF, PANS, weather, and external intelligence into one maritime entity.
Evidence-first alerts
Every alert carries track history, geofence events, transcripts, documents, EO frames, or registry matches.
RAG MDA assistant
Searches vessel history, PANS records, VHF transcripts, reports, and intelligence records with cited context.
Advisory-mode AI
AI recommends and explains while final watchkeeping decisions remain under authorised human control.
Architecture
Retrofit software layer. Defence-grade operating posture.
01
Sensor and external feed adapters
02
Ingestion, normalisation, and common maritime model
03
Radar/AIS/EO association and vessel entity fusion
04
Rules, geofences, anomaly models, and risk scoring
05
Evidence store, replay, audit, and semantic search
06
Watchkeeping portal, VOI ranking, and MDA assistant
Core stack
Kafka or Redpanda, MQTT, PostGIS, TimescaleDB, ClickHouse, OpenSearch, vector search, YOLO or DETR, Whisper or Indic ASR, LLM extraction, MLflow, Kubernetes, Prometheus, Grafana, RBAC, encryption, and audit trails.
Pilot path
From proposal to controlled operational validation.
010-5 months
Foundation and offline prototype
Acquire sample data, define schemas, build ingestion, implement baseline fusion, geofence logic, risk scoring, and an initial dashboard.
024-9 months
AI modules and validation
Add anomaly detection, behavior analytics, EO classification, PANS extraction, VHF ASR/NLP, error analysis, and false-alert tuning.
038-14 months
Controlled live deployment
Integrate selected live feeds in advisory mode, compare outputs with watchkeeper observations, and build model governance.
0414-18 months
ICG collaboration and scale-out
Run UAT, refine operational workflows, harden security, prepare V&V evidence, training material, and deployment guidance.
Validation metrics
Built to be measured in the pilot, not merely admired in a deck.
The demonstration target is a pilot-ready maritime decision-support system with measurable association accuracy, false-alert reduction, latency, extraction quality, and operator workload impact.
AIS-radar association accuracyDark vessel detection precisionEO classification accuracyPANS extraction qualityVHF keyword/entity accuracyFalse alert reductionOperator workload reductionSensor event to alert latency
Prototype scenario
A clean narrative for demo, validation, and stakeholder review.
- Simulated AIS and radar feeds enter the ingestion layer.
- A radar contact continues without an AIS match.
- Another vessel enters a restricted geofence and begins loitering.
- EO classification conflicts with the AIS-declared vessel type.
- PANS data shows a missing or mismatched pre-arrival record.
- VHF transcripts reveal suspicious meeting-related communication.
- The risk engine ranks the vessel HIGH with evidence attached.
- The MDA assistant summarises why a watchkeeper should investigate.
Secure by design
Deploy as an on-prem or private-cloud intelligence layer.
AMSAS is shaped for coastal surveillance environments where audit, explainability, role-based access, model traceability, and operator trust matter as much as detection speed.