Home / Platform
One reasoning layer. Six disciplines.
Port-Systems isn't a chatbox bolted onto a database, and it isn't a rigid rules engine. It's a neuro-symbolic stack: connectors, knowledge graph, ontology, rules, a neural query interface, and the outputs your business actually consumes.
The architecture
From raw sources to formally-derived answers
Each layer has a distinct job. The neural layer understands; the symbolic layer reasons; the governance layer proves. Skip any and the answers stop being defensible.
Layer 06
Aggregation
Connectors
APIs, databases, log files, spreadsheets, supplier portals, ingested without forcing teams off the tools they already use.
Layer 05
Knowledge graph
Symbolic representation
Vessels, berths, meters, contracts, assets and people, modelled as typed entities and relationships. The estate as a graph the platform can actually reason over.
Layer 04
Domain ontology & rules
Symbolic logic
Carbon factors, marine taxonomies, tariff structures and operational policies expressed as versioned, reviewable rules, not statistical patterns.
Layer 03
Neural query interface
Language to logic
A language model interprets the user's intent and compiles it into a formal query against the graph. The LLM never invents the answer, it only frames the question.
Layer 02
Governance
The data boundary
Role-based access, full lineage and local-first guardrails. Every derivation is logged with its inputs, rules applied and source records.
Layer 01
Output
Recurring reports & insight
Monthly board packs, ESG submissions and ad-hoc questions, each answer carrying the symbolic derivation behind it.
Why hybrid
Neural reach with symbolic accuracy.
Pure LLM tools hallucinate because they generate plausible-sounding text. Pure rules engines can't cope with the variety of estate data. The neuro-symbolic stack uses each for what it is good at, and nothing more.
Trade-off
Neural alone
Reads everything as language. Confident, fluent, and often wrong. No audit trail.
Trade-off
Symbolic alone
Precise but rigid. Every question needs a developer. Every new source needs a rewrite.
Our approach
Neuro-symbolic
The model understands the question. The graph and rules derive the answer. Both are inspectable.
Day in the life
Same platform. Two very different users.
Port-Systems serves the floor and the boardroom from the same graph, what each role sees is governed by access, not by buying a separate product.
Operational user
Estates Manager
Show me energy spikes in South Harbour yesterday.
Anomaly at 14:00. Shore power load exceeded baseline by 15% (rule: ENERGY.SPIKE.σ>2).
Cross-reference with vessel arrivals.
Graph joins to two cruise vessels berthed 13:50–18:30. Citing AIS log #221, Meter A4.
Strategic user
Finance Director
Draft the monthly carbon report for the Board.
Compiling derivation. Scope 1: fleet rule applied. Scope 2: half-hourly meter aggregation.
Include source citations.
Each row links to Invoice #4401, Meter Log #B2, Vessel Log #M03 and the rule version used.
Implementation
From signed-off scope to live monthly outputs in 12 weeks
Each node ships in a fixed three-month sprint: ontology, rules and outputs delivered together. We pilot in parallel with your existing reporting so nothing breaks.
Weeks 1–2
Co-design
Confirm use cases, identify graph entities, lock KPIs.
Phase 1
Weeks 2–4
Discovery
Map sources, draft the domain ontology, agree security protocols.
Phase 2
Weeks 4–8
Build
Ingestion pipelines, knowledge graph build-out, rule authoring.
Phase 3
Weeks 8–10
Pilot
Deploy, train champion users, run parallel to existing reporting.
Phase 4
Weeks 10–12
Evaluation
Audit derivations, measure savings, plan the next node.
Phase 5