A major Canadian airline
As an employee of a consultancy, on the client's programme
A 25+ API micro-service architecture carrying a 1,000 req/sec SLA, in front of reservation systems that do not forgive sloppy integration.
- Amadeus Altea
- Sabre
- Azure Service Bus
- Micro-services
- DLQs and retry scopes
Context
A major Canadian airline, delivered by a consultancy I worked for. Neither is named here. What is publishable is the shape of the work: enterprise integration at a contractual throughput SLA, against reservation systems that were designed decades before the services calling them — and my part in it, which was contributing to the architecture and building services under it rather than owning the engagement.
The problem
Airline reservation systems are the canonical hard integration target. They are authoritative, they are old, they have their own availability characteristics, and a customer-facing booking flow has to stay responsive regardless of what any of them is doing at that moment.
The 1,000 req/sec figure was not an internal aspiration — it was written into the client's agreement with us. At that number there is no room for the usual escape hatch of retrying synchronously and hoping. Back-pressure has to be designed in, so a slow upstream degrades throughput in a controlled way rather than cascading into timeouts across twenty-five services.
What I built
- Micro-services across the booking path
- Booking, seat maps, flight search and automated baggage handling, each with its own failure characteristics and each held to the same latency budget.
- Integration against Amadeus Altea and Sabre
- Against the reservation systems directly, with the mapping and error semantics that implies.
- Back-pressure and failure isolation
- Queue-based decoupling with retry scopes and dead-letter queues, so a degraded dependency is contained rather than propagated.
Outcome
The system went live and met its 1,000 req/sec SLA under test and in production. I contributed to the architecture and to the services beneath it; the engagement belonged to my employer, and the parts of it I can speak to are the ones I built.
Sound like something you have?
More work
- KeystoneOS
Idempotent webhook ingestion keyed on provider event IDs, anonymous-to-identified event backfill, CREA RESO OData sync, and pgvector semantic search over listings.
- Surefire Fundraising
A Laravel platform handling catalog, checkout and transactional email for an Edmonton client whose business was built on top of it.
- What's The Deal
A Flutter app on the App Store and Google Play using location, geofencing, maps and push, with RAG-based search and a vector recommender behind it.