Business workbenches
Procurement buyer, fleet operations, ESG, legal, finance and compliance screens with human approval and audit trail.
Technical architecture / From workflow concept to deployable AI capability
AI solution blueprint for Berge Bulk.
The business workflows show what users would experience. This page explains what must exist behind them: system integrations, data products, AI services, workflow controls, governance, security and delivery work.
01 / Target architecture
Copilots and agents
Recommendation services, document intelligence, retrieval, reasoning prompts, deterministic calculators and explainability modules.
Curated operational data
Reusable data products for requisitions, suppliers, vessel performance, voyages, fuel, carbon, documents and claims.
APIs, events and connectors
Connectors to SERTICA/Moscord, PMS, voyage systems, AIS/weather, ERP/finance, DMS/email and identity services.
Security, MLOps and controls
Access control, model monitoring, prompt/version management, policy guardrails, feedback capture and compliance logging.
02 / System integrations
System / source
Data needed
AI use
Write-back
SERTICA / Moscord
Requisitions, RFQs, PO lines, inventory, supplier quotes, delivery status
Supplier ranking, price anomaly detection, spares demand forecast
Approval note, risk tag, suggested PO split, invoice exception
PMS / technical systems
Equipment, maintenance schedule, defect history, alarms, criticality
Part-fit matching, urgency scoring, failure-risk context
Recommended spare bundle, technical confirmation task
Voyage optimisation / AIS / weather
Route, ETA, speed, weather, sea state, port congestion, berth window
Fuel and ETA prediction, multi-objective route/speed optimisation
Captain instruction, operational exception, ESG report input
Noon reports / vessel performance
Fuel consumption, RPM, draft, trim, engine load, hull performance signals
Baseline performance modelling, anomaly detection, saving verification
Performance review summary, variance explanation
DMS / email / contracts
Charterparties, SOF, NOR, invoices, survey reports, port documents
OCR, clause extraction, evidence retrieval, claim and risk generation
Evidence pack, claim draft, dispute note, compliance task
ERP / finance / ESG
Costs, invoices, payments, claims, bunker prices, carbon factors
ROI calculation, cost allocation, carbon-cost reporting
Approved adjustment, accrual input, customer emissions report
03 / Data products to build
Supplier performance mart
Supplier, part category, port, price history, lead-time reliability, rejection rate, invoice exceptions and warranty outcomes.
Vessel equipment graph
Vessel, equipment, serial number, drawing revision, compatible part, maintenance event, alarm pattern and criticality relationship.
Voyage performance mart
Voyage, route, weather, speed, consumption, ETA, berth window, port delay, hull condition and realised savings.
Carbon calculation dataset
Fuel type, emissions factor, ETS/FuelEU assumptions, voyage allocation, customer reporting fields and audit version.
Document evidence index
Parsed clauses, entities, timestamps, cargo details, invoice lines, SOF events, citations and confidence scores.
Decision outcome log
AI recommendation, human decision, override reason, realised result, feedback and model improvement signal.
04 / AI facilities and their roles
Grounded answers and generated packs
Retrieves approved policies, contracts, manuals and evidence before generating PO notes, voyage instructions or claim drafts.
Forecast risk and value
Predicts supplier lead time, purchase anomaly, fuel consumption, ETA risk, claim value and likely evidence confidence.
Recommend the best trade-off
Balances cost, lead time, off-hire risk, bunker saving, carbon exposure and berth constraints using deterministic business rules.
Extract operational evidence
Applies OCR, layout parsing, entity extraction, table extraction and clause matching to voyage file packs.
Keep decisions auditable
Prevents unsupported claims, enforces citation requirements, routes high-risk actions to human approval and logs model versions.
Improve from adoption
Captures accepted, rejected and overridden recommendations so models can be tuned against real operational outcomes.
05 / Security, governance and operating model
Identity and access
Use SSO/RBAC so procurement, operations, finance and legal see only permitted vessels, vendors, contracts and documents.
Human approval
AI can recommend and draft, but PO release, captain instruction, claim submission and compliance clearance require named human approval.
Audit trail
Store prompt version, data snapshot, source citations, model output, human decision and realised business result.
Data protection
Classify contracts, supplier terms and operational documents. Mask sensitive fields in lower environments and restrict model training reuse.
Model monitoring
Track hallucination rate, citation coverage, recommendation adoption, override reasons, realised ROI and drift in supplier/voyage patterns.
Change management
Define process owners, model owners, escalation paths and monthly AI value review with business and IT stakeholders.
06 / Delivery roadmap
Discovery and data access
Confirm systems, owners, sample data, security constraints and value metrics for one selected pilot.
Data foundation and prototype
Build connectors, curated data product, retrieval index and a clickable workflow with historical back-test.
Controlled pilot
Deploy to a limited user group, keep human approval, measure adoption, cycle time, accuracy and realised value.
Scale and industrialise
Harden APIs, monitoring, audit, security, training, support model and reusable AI services across the three value pools.