6. Implement Text Analysis Solutions (10-15%)

The final content domain completes the pattern you saw in vision: an LLM-first reframing. Classification, sentiment, entity extraction, and summarization — once the exclusive province of dedicated NLP endpoints — are now often handled by a generative model with a good prompt and structured output. But the dedicated Azure Language service hasn't vanished; it remains the right answer for high-volume, low-latency, PII-sensitive, and specialized cases. The exam tests whether you can choose correctly between the two, and constrain the model's output when you go LLM-first.