App Integration, Analytics & AI Services
SQS, SNS, EventBridge, Kinesis, Athena, Glue, QuickSight, and the AI service family — the final sweep of Domain 3 services.
Application integration (decoupling)
Decoupling means components communicate through an intermediary instead of directly, so one component's failure or slowness doesn't break the others. Two services dominate this space — and one exam distinction.
Message queue: producers drop messages, consumers pull them when ready. One consumer processes each message. Decouples and buffers workloads.
Pub/sub notifications: one message is pushed to many subscribers at once (email, SMS, Lambda, SQS queues).
Serverless event bus routing events between AWS services and SaaS apps based on rules.
Visual workflow orchestration of multi-step processes with branching, retries, and human approvals.
SQS = queue, pull, one consumer ("buffer requests between web tier and workers"). SNS = topic, push, many subscribers ("notify multiple systems the moment an order ships"). "Fan-out" = SNS topic feeding multiple SQS queues.
Analytics
Collect and process real-time streaming data (clickstreams, IoT telemetry) as it arrives.
Serverless SQL queries directly on data in S3 — pay per query, no infrastructure.
Serverless ETL (extract, transform, load) and data cataloging to prepare data for analytics.
Serverless BI dashboards and visualizations.
Managed big-data frameworks (Spark, Hadoop) for massive processing jobs.
Search and log analytics engine.
"Query data already in S3 with standard SQL, serverless" → Athena. "Real-time streaming" → Kinesis. "Prepare/transform data" → Glue. "Visual dashboards for business users" → QuickSight.
AI & ML services (know the one-liners)
Build, train, and deploy custom ML models.
Image and video analysis (faces, objects, moderation).
Natural language processing: sentiment, entities, PII.
Text to speech.
Speech to text.
Language translation.
Chatbots (the tech behind Alexa).
Extract text and data from scanned documents/forms.
Intelligent enterprise search.
Recommendation engines.
Time-series demand forecasting.
Build generative AI apps using foundation models via API.
These appear as one-line matching questions. Watch the easy trap: Polly speaks (text→speech), Transcribe listens (speech→text).
An architecture needs a buffer between a web front end and worker processes so traffic spikes don't overwhelm the workers. Which service fits?