Case studyAI · Maritime intelligence

AEGIS

A multi-agent platform for monitoring maritime shipping risk across chokepoints, supply chains, and tariff regimes — built on top of seven live data sources, four database technologies, and a CrewAI orchestrator with W3C PROV lineage.

7+Data sources

4Databases

3LLM agents

W3C PROVLineage layer

✓build aegis --release~/aegis9 deps0

Ready Stack resolved in 0.23s
[+] loaded dependency: FastAPI
[+] loaded dependency: React 19
[+] loaded dependency: CrewAI
[+] loaded dependency: PostGIS
[+] loaded dependency: Neo4j
[+] loaded dependency: ChromaDB
[+] loaded dependency: Redis
[+] loaded dependency: deck.gl
[+] loaded dependency: Docker

01 / Operations views.

Three views, one mental model. The 3D globe is for situational awareness; the 2D map for incident triage; the dashboard for daily standup metrics.

**Operations dashboard**Vessel state, risk heat-map, and ATA forecasts in one pane.

**3D globe view**Live AIS positions overlaid with active risk zones and choke-point density.

**2D operational map**Standard mercator projection for incident review and routing decisions.

02 / AI analyst.

Natural-language queries route to a CrewAI orchestrator with three specialised agents — risk, routing, compliance. Every intermediate step is captured with W3C PROV lineage; final answers ship with structured citations and a hallucination check pass.

**Analyst — Query**Natural-language query → CrewAI delegates to risk, routing, and compliance agents.

**Analyst — Reasoning**Each agent's intermediate output is captured with W3C PROV lineage for auditability.

**Analyst — Results**Synthesised answer with structured citations, source data, and hallucination check.

03 / Risk analytics.

Choke-points (Strait of Hormuz, Suez, Malacca) are weighted by marine weather, active risk zones, and tariff exposure. Each variable is independently queryable, then composed into a single per-vessel risk score.

04 / Architecture.

7 ingestion sources land in a Redis cache, then split across PostGIS (vessels & zones), Neo4j (route + supplier graphs), and ChromaDB (vector store for the analyst's RAG layer). FastAPI sits in front; deck.gl renders globe + map; CrewAI runs the agent loop on a separate worker.

System architecture diagram — **System architecture**End-to-end pipeline from ingestion → storage → orchestration → presentation.

Data flow diagram — **Data flow**How each source's records get normalised, joined, and indexed.

Entity-relationship diagram — **Entity-relationship**Vessel, route, port, supplier, risk-event — and how they link.

API documentation overview — **API surface (1/2)**OpenAPI-generated Swagger for the FastAPI service.

API documentation detail — **API surface (2/2)**Agent endpoints, with request/response schemas inline.