Data quality agent
Finds missing fields, cart attribution gaps, offline records, weak signal, and low-confidence evidence before reports are trusted.
Internal GCCAP operating surface
This page is retained behind the public client website.
GCCAP does not delete operating records, trial packs, hardware activation surfaces or evidence automation layers. They are preserved for authorised client portal, command-centre and internal operating use while the public website stays clean for airline first impressions.
Open portal accessInternal GCCAP surface
AI agents that assist GCCAP without taking uncontrolled action.
Internal system phase introduces controlled AI-style operational agents that read the GCCAP evidence chain, draft recommendations, flag risks, prepare founder briefings, and support report review while keeping all high-risk outputs under human approval.
StatusInternal / restricted surface. Not part of the public client walkthrough.
GCCAP
Sensor
Gateway
Evidence
Watchdog
Journey intelligenceRepresentative demo
Agent postureRecommendation-only
Approval modelHuman review required
Public safetyWatchdog remains blocked
Operational layer
What Build 10 completes
The AI layer now sits above the command centre and evidence chain. It is designed to increase speed and quality without creating unsafe automation or unsupported public claims.
Journey reconstruction agent
Reviews stage windows, soak coverage, dry ice process evidence, truck dwell, aircraft handover, and replay exceptions.
Breach analysis agent
Summarises serious incidents, likely causes, evidence confidence, and corrective-action priorities for human review.
Report drafting agent
Drafts client-facing narrative inputs but cannot approve or issue reports.
Device health agent
Flags low battery, weak signal, silent sensors, gateway issues, and assignment gaps.
Watchdog review agent
Blocks public use until sample thresholds, aggregation, anonymisation, confidence, and legal review are complete.
Founder briefing agent
Turns current system state into a practical operating brief: priorities, risk, money, trust, and next move.
Decision clarity
AI safety boundaries
Build 10 treats agents as assistive operators, not autonomous authorities. This protects GCCAP from overclaiming and preserves trust with airlines, caterers, partners, and the public.
AllowedGenerate recommendations, draft report notes, flag review items, produce founder briefings, and recommend actions.
BlockedNo public Watchdog publishing, no client messaging, no report issuing, no live rule changes, no data deletion, and no incident closure without human approval.
AuditEvery run, recommendation, approval decision, and configuration change is recorded in the agent audit trail.
Production statusPilot-stage local engine; later production builds must add model-provider controls, prompt/version management, evaluation tests, cost controls, and secure tool permissions.
Data path
Where AI sits in the GCCAP evidence chain
Agents consume outputs from intake, pipeline, journeys, rules, reports, portal, alerts, and command centre. They do not replace the source evidence chain.
01Sensor captures cart temperature behaviour
02Gateway collects sensor records
03Raw payload is received and preserved
04Payload is mapped to sensor, cart, gateway, site and client
05Temperature event is normalised into the GCCAP event model
06Event is assigned to the correct cart journey stage
07Rules engine flags warning or review conditions
08Recovery and duration are calculated
09Data completeness and hardware health are assessed
10Evidence report is prepared for private client review
11Approved outputs can support client action and future benchmark methodology
Strongest next move
The next build is Watchdog governance
After controlled AI recommendations, GCCAP should build the private-public governance layer that determines what can be aggregated, anonymised, confidence-rated, legally reviewed, and eventually published.
Request a controlled pilot