Train on real backend problems.
API Observability with Prometheus
Instrument a running API until one dashboard answers what is failing, how badly, and since when — then break it on purpose and find out whether the answer was true.
Event-Driven Order Pipeline
Rebuild a slow, synchronous checkout around a message queue — accept orders fast, process them async, lose nothing.
The Query That Got Slow
An endpoint that was fast at a thousand rows crawls at a million. Read the plan, fix it, and prove which change was the one that mattered.
Two Writers, One Row
Concurrency does not announce itself in tests that run one request at a time. Oversell the inventory on purpose, then make it impossible.
Stop Writing Auth: Delegate to Keycloak
Hand identity to an OIDC provider and defend the boundary that is left: validating tokens you did not issue, and surviving a key rotation.
A Serverless Read Path That Survives a Cold Cache
API Gateway, Lambda, DynamoDB and a Valkey cache. Make it fast, then take the cache away mid-flight and find out what you actually built.
The Dashboard That Stayed Green
For eighty minutes 4% of charges failed while every panel stayed green: the failures were fast and they returned HTTP 200. Build the instrumentation that would have gone red — and prove it under the conditions that fooled the last one.