Contract the output
Define the shape the application can accept, validate every response, and fail closed when the model returns something outside it.
CITYJS ATHENS 2026
How to Become an AI Engineer
Building Reliable AI Features in TypeScript
AI engineering begins after the first successful model call. Build a TypeScript feature with schema-validated output, evaluations, evidence-based model selection, production tracing, and a release gate.
CityJS CityAI Athens · 21–23 October 2026
From first model call to release gate.
The workshop follows one TypeScript feature through the controls that make its behavior testable, explainable, and safe to change.
Evaluate every change
Put representative cases, deterministic assertions, and calibrated grading into CI so prompt and model changes have evidence behind them.
Engineer the context
Treat retrieval, compaction, memory, cost, and latency as deliberate system-design choices rather than incidental prompt details.
Select and route
Compare models against the feature's real workload, then route each task by measured quality, cost, and latency.
Trace and investigate
Record each call so a production failure leads to its trace, its cause, and a new case in the evaluation suite.
Gate the release
Require schema-valid output, green evaluations, acceptable cost, and production tracing before a change ships.
Who it is for
TypeScript engineers building AI features that need to move from plausible demos to dependable product behavior.
Comfortable with TypeScript
Experience with model APIs
No machine-learning background needed
Leave with evidence, not just a demo.
The feature is schema-valid, evaluated, inside its cost budget, routed, and traceable. The next prompt or model change must pass the same release gate.