Build architectures from blocks and test them against real traffic patterns
Serve a marketing website with global reach. The site is fully static — prebuilt HTML, CSS, JS, and images, no app server or database needed. Drop the assets into object storage so S3 alone can serve them, then put a CDN in front to handle the load cheaply and keep latency low worldwide.
Design a basic web application that can handle moderate traffic. Route requests through a load balancer to web servers backed by a database and cache.
Build a read-heavy API that serves product data. The database alone can't handle peak load — introduce a caching layer to absorb repeated reads and protect your database.
Build a document upload service where every file must be validated, deduplicated, and stored. A Step Functions state machine orchestrates the Lambda pipeline so the API responds fast while processing happens serverlessly — no servers to manage, just workflow and storage.
Build a mobile push-notification service. When an event fires, a Step Functions workflow triggers Lambda to prepare the message, then Lambda publishes it to SNS, which fans notifications out to hundreds of thousands of subscribers.
Handle a flash sale for an online store. Traffic spikes 10x during the sale window. Requests involve browsing (reads), adding to cart, and checkout (writes). The system must stay up during the spike or you lose revenue.
Build an image upload service where users submit images that need background processing. The API must respond quickly while heavy processing happens asynchronously via a message queue.
Power a product search feature that handles autocomplete and full-text queries. A relational database is too slow for search — use a dedicated search engine and cache hot queries for sub-100ms responses.
Build a chat service where users hold long-lived connections and broadcast messages to rooms. A streamer goes live midway and a flood of viewers joins. You need WebSockets, pub/sub fan-out, and history persistence — all under 200ms.
A fleet of 30,000 IoT sensors streams readings every second. Latency to the device is loose, but the write throughput is relentless and the dashboards expect fast range queries. Decouple the ingestion path, aggregate over windows, and persist to storage built for time-series.
Build a checkout pipeline for an e-commerce platform. Orders arrive through the CDN, are authorized by the API gateway, and a Step Functions workflow coordinates serverless fulfillment steps. Order data lives in DynamoDB, which scales reads natively — no cache needed. The checkout occasionally gets hammered by bot traffic, and the state machine throttles mid-event.
Build a clickstream analytics ingestion pipeline. Client events arrive at the API, a Step Functions workflow validates and routes them into a queue, and workers transform the data into a S3 data lake. Writes dominate — the pipeline must keep up with spikes without dropping events.
Build a College Event Registration System on AWS. Students access a website, register for an event, upload documents, and receive a confirmation notification while organizers monitor the application. Serve the site through CloudFront, process registrations serverlessly, store them in a fast NoSQL store, keep uploads in object storage, and confirm via a notification service. You do NOT need every service — choose only what the scenario requires.
Design the backend for a social media feed that serves personalized timelines. Read traffic is 50x write traffic. A viral post causes a sudden spike. You need caching, a database, and enough compute to survive the surge.
Build a globally distributed API that serves mobile and web clients across regions. Traffic comes in waves as different time zones wake up. The API must route through security layers, manage rate limits, and keep latency under control worldwide.
Build a video platform where creators upload footage and viewers stream it globally. Uploads are orchestrated by a Step Functions workflow that fans transcode jobs into a queue for workers; finished renders land in S3 and stream through the CDN. Catalog data lives in DynamoDB. The broker loses a partition at t=55 — plan for it.
Build a real-time fraud detection service for a payments platform. Every transaction is scored: the Step Functions workflow coordinates the Lambda scoring step, results are stored in DynamoDB, a stream processor flags anomalies, and every event is archived to S3. Attackers probe the API in waves — the WAF is the first line of defense.