AI App Engineering Lab

09:00 – 10:15

See what a model call really is

—the modern AI app stack, tokens and context, model selection trade-offs

—Takeaway Three models compared on one task — latency, cost and quality on one table you made.

10:30 – 11:45

Design prompts and messages like an engineer

—system vs user intent, message design, prompt assets vs chat

—Takeaway A prompt rewritten three ways with the output diffed — the vague version retired for good.

12:00 – 13:00

Constrain the output with a schema

—structured outputs, validation, retry-on-invalid

—Takeaway Model responses bound to typed, validated objects — a rename breaks the build, not production.

14:00 – 15:15

Stream it into a real interface

—streaming responses, conversation state, error handling

—Takeaway A streaming AI endpoint in your own application, surviving a dropped connection.

15:30 – 17:00

Build the first working feature

—wiring it together, the happy path, the first failure path

—Takeaway Your first end-to-end AI feature running — small, structured and testable.

09:00 – 10:15

Give the model tools

—function calling, tool schemas, external APIs

—Takeaway The model completing a task by calling your functions — with the call log read aloud.

10:30 – 11:45

Constrain what it may do

—boundaries, human approval, side effects

—Takeaway A consequential action gated behind explicit approval — demonstrated by refusing one.

12:00 – 13:00

Handle failure like production

—rate limits, retries, timeouts

—Takeaway The feature surviving a provider outage, injected live — degrading, not dying.

14:00 – 15:15

Control the cost

—token budgets, caching, cost ceilings

—Takeaway A cost ceiling that actually stops spend, proved against a runaway loop.

15:30 – 17:00

Integrate, demo and defend

—end-to-end integration, the demo, design review

—Takeaway Your feature demoed end to end and reviewed — with the design notes you take home.