Automation case study
API Integration Platform
A middleware layer connecting CRM, billing, email and internal tooling behind one consistent API.
- New integration time
- -70%
- Silent failures
- → 0
- Delivery success
- 99.4%
200 OK · 142 ms
{
"queued": 42,
"idempotent_skips": 3,
"dead_letter": 0
}Interface reference — rendered from the real component structure, not a stock image.
The problem
Five services each had their own auth, payload shape and failure mode, and every integration was rewritten from scratch.
The solution
One normalised internal API with adapters per provider, shared retry and backoff, and dead-letter handling for anything that keeps failing.
Hard parts
Partial failures across providers. Each operation is idempotent and safely replayable from the dead-letter queue.
Outcome
New integrations now take days rather than weeks, and failures are visible instead of silent.
Technology used
- Node.js
- Express
- TypeScript
- Redis
- Docker
How it was delivered
- 01Client problemEvery new tool meant bespoke glue code nobody wanted to own.
- 02ResearchCatalogued each provider's auth model, rate limits and webhook guarantees.
- 03PlanningDefined one internal contract; providers became swappable adapters.
- 04WireframeSequence diagrams for each critical flow before implementation.
- 05DevelopmentExpress service, Redis-backed queue, structured logs with correlation IDs.
- 06TestingContract tests per adapter plus chaos runs with forced provider errors.
- 07DeploymentDockerised, deployed with health checks and per-queue alerting.
- 08ResultsIntegration work became routine instead of a project.
Lessons learned
Idempotency keys everywhere. They cost nothing on day one and save the weekend later.