Microsoft · AI-103
Optimize and operate generative AI
AI-103 · Lesson 9 of 15 · 1:05:27
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the AI-103 exam (Azure AI Apps and Agents Developer Associate). In this lesson: prompt engineering and model parameters (system messages, few-shot, temperature and top_p, output limits, structured outputs), reasoning models (reasoning effort, hidden reasoning tokens, summaries), chain-of-thought and maker-checker self-critique loops, observability (OpenTelemetry tracing, token analytics, latency), and orchestrating models and rules (model router modes, one model per agent, deterministic routing). Includes three original practice questions.
Preview the topics
- A ticket classifier, on a model without reasoning, returns different labels for the same ticket from one run to the next. Its temperature is set to 1.0. What should you change first?
- On hard questions, a reasoning model returns responses with the status incomplete, and sometimes no text at all. What should you change?
- A legal team sends requests through the model router, in its default mode. For contract reviews, they want the best possible answers, whatever the cost. What should they change?