Comprehend

Pretrained NLP tasks vs Comprehend Custom for domain-specific text, and how real-time calls differ from asynchronous batch jobs.

Detect sentiment, language, entities, and key phrases in a block of text via SDK - five basic and two intermediate examples in Python and TypeScript.

Pulling structured signals out of unstructured text with DetectEntities and DetectKeyPhrases - types, offsets, scores, and combining both in one pipeline.

Multi-language sentiment analysis via SDK - DetectSentiment's Positive/Negative/Neutral/Mixed scoring and using DetectDominantLanguage to route text first.

Choosing synchronous Detect calls vs asynchronous Start*Job batch analysis - latency, cost, and IAM shape for each, and how to poll a job to completion.

Language coverage, batching strategy, and PII-handling defaults - a checklist for production Comprehend use across boto3 and AWS SDK v3.