Keyword → service: the mapping drill
Lesson 2 of 4 in Exam Readiness: Reading Questions Like an Engineer.
Fluent test-takers do not deliberate on every item — they recognize. Exam scenarios are written in a controlled vocabulary, and certain phrases are near-deterministic pointers to a service. Build these as reflexes: see the phrase, name the service, then verify against the qualifier. The table below is the drill sheet; the classifier underneath is the test.
| The scenario says… | The exam wants | Why |
|---|---|---|
"sub-second latency at unpredictable scale" | SageMaker real-time endpoint + auto scaling | Persistent low-latency serving that follows load |
"intermittent traffic, tolerates cold starts, no instances to manage" | SageMaker Serverless Inference | Pay-per-use serving with zero capacity management |
"score the whole dataset nightly, no persistent endpoint" | Batch transform | Bulk offline inference as a job |
"generative app, no infrastructure to manage" | Amazon Bedrock (serverless FM API) | Hosted FMs, pay per token |
"reduce hallucination with company documents" | Bedrock Knowledge Bases (RAG) | Grounds answers in retrieved sources |
"block harmful prompts / filter PII in responses" | Bedrock Guardrails | Policy filters on both input and output |
"catalog and crawl data in S3" | Glue crawlers + Data Catalog | Schema discovery into the shared metastore |
"ad-hoc SQL directly on S3" | Amazon Athena | Serverless queries, no loading |
"join streams / exactly-once / windowed aggregation" | Managed Service for Apache Flink | Stateful stream processing |
"easiest way to land a stream in S3" | Amazon Data Firehose | No-code buffered delivery |
"extract text, tables, and forms from scanned documents" | Amazon Textract | Purpose-built document AI |
"multi-step ML workflow with retries and branching" | Serverless state machines with SageMaker/Bedrock integrations | |
"human review of low-confidence predictions" | Amazon A2I (Augmented AI) | Managed human-review loops on model output |
"share features across teams, online and offline" | SageMaker Feature Store | One feature pipeline, both read paths |
"track and compare experiment runs" | SageMaker managed MLflow | Experiment tracking without running a server |
"audit who called which API" | AWS CloudTrail | The API audit record |
Drill it like flashcards: cover the right-hand columns, read the phrase, answer out loud. When a reflex misfires twice, go reread that service's module — a wrong reflex is worse than none, because it feels like knowledge under time pressure.
Interactive sorting exercise: Sort each scenario phrase onto the service the exam wants.
Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.