01 The challenge
The airport operator's flight-settlement application sat at the heart of the airport's revenue operation: it reconciled what each airline owed for landings, parking, passenger services, and ground handling. The software had been written a decade earlier. The code was obsolete, integrations were manual, and manifest data — the source of truth for every settlement — kept arriving as PDFs that humans re-entered into the system.
At airport scale, that workflow was an operational risk. Every manifest processed by hand was a chance for a data error, a late month-end close, or an airline dispute. The operator needed more than a refactor. They needed a re-engineered application that could automate manifest ingestion, settle faster, and scale toward full API integration with airlines.
02 What we built
The engagement ran on the ITSense Method — Sense to Operate — with the Aviation Cell. The substrate was active from day zero: Claude Cowork held the shared context; MCP gave AI agents access to the existing codebase and data warehouse; every stakeholder meeting at the operator was captured by the intelligence pipeline.
Specifically, we delivered:
- Total re-engineering of the flight-settlement core, modernizing the data model and breaking the coupling to legacy storage.
- OCR + AI for manifest extraction, trained on the operator's historical PDF corpus and validated against real-time operations.
- Automatic alerts for exceptions — a manifest that doesn't match a scheduled flight, a settlement out of tolerance, a document with degraded OCR confidence.
- A code reorganization that left the application maintainable by the operator's own team, not in eternal dependency on ITSense.
- A roadmap to direct API integration with airlines, eliminating the need for physical paperwork over time.
Tech stack
03 The outcome
The new system reached 97% precision on OCR-driven manifest extraction in production — the threshold where human review becomes the exception, not the rule. Settlement cycles compressed. The operator's ops team moved from typing data to reviewing flags.
04 What it meant
"This project is a milestone of technological innovation for our operator, consolidating us as a leader in the adoption of technology solutions in the airport industry, while significantly improving the experience of our users and customers."— South America's leading airport operator (NDA client)
Beyond the operational metrics, the project reset how the operator thinks about software: away from long, risky refactors and toward an AI-native delivery method where systems are re-engineered in shippable weekly increments, with AI in every loop and humans owning every irreversible action.
05 Team & Method
Aviation Cell — cross-functional engineering, data, OCR, and airport-operations domain experts.
Every phase of the ITSense Method was active. Sense — the team ingested the legacy codebase and historical manifests via MCP; Claude produced the first brief; humans questioned and refined. Shape — ADRs co-produced, specialist AI agents reviewed security, data, and FinOps. Forge — shippable weekly increments, AI pair-programming on every task, AI review before human review on every PR. Prove — AI-generated tests plus adversarial testing on the OCR pipeline. Operate — production from week one with observability in place.
Every production deploy, schema change, and new model version required explicit human approval under the Human Oversight Protocol. AI proposed. Humans disposed.
