How CloudSynth works
The core loop — read, write real CDK, run synth & validate — and the philosophy behind it.
CloudSynth is built on one belief: in the age of AI-generated infrastructure, the scarce skill isn't writing code — it's knowing whether the code is right.
The philosophy
AI will happily generate a CDK stack for you. It will also happily generate one with a public bucket, an over-permissive IAM role, or a property that silently doesn't do what you think. The engineers who thrive now are the ones who can generate quickly and catch what's wrong — and that judgment only comes from writing real infrastructure and having it graded honestly.
So CloudSynth never grades by comparing your code to a stored answer. There are many correct ways to write a stack. Instead, we synthesize your code into the same CloudFormation AWS would receive, and assert on the properties of the result — is the bucket versioned, is public access blocked, does the role follow least privilege. Your style is yours; the architecture is what's checked.
Use AI while you learn — really. Lessons assume you have an assistant open. The point isn't to write CDK unaided; it's to develop the eye that catches what your assistant gets wrong. The engine gives you the honest feedback loop to build that eye.
The core loop
Every exercise checkpoint follows the same rhythm:
- Read — a short explanation of the concept, with the reasoning, not just the syntax.
- Build — you complete real CDK code in the editor, in TypeScript or Python.
- Run synth & validate — your code runs through the four-stage engine below.
- Iterate — failed assertions come with hand-written hints; fix and re-run until it's green.
When every assertion passes, the checkpoint is complete — recorded server-side, so your progress reflects verified work, not self-reporting.
What counts toward a concept's progress
A concept's progress bar counts every lesson in it, not just the checkpoints — reading lessons count too, including the "deploy it for real" lesson at the end of a concept. Reading lessons don't have anything to verify, so they're marked complete the simple way: click Continue (or Back to concepts, on the last one) at the bottom of the page. That's a record of how far you've read, including on the deploy lesson — clicking Continue there confirms you worked through the steps, not that a deploy actually succeeded.
Checkpoints are the only lessons held to the verified standard described above. If a concept's progress looks stuck short of 100%, it usually means one reading lesson still hasn't had its Continue button clicked — often the last "deploy it for real" one, easy to miss since there's no code to run on that page.
What "synth & validate" runs
Your submission executes in an isolated sandbox — it never touches a real AWS account and holds no credentials. Four stages run in order:
cdk synth does.Everything you see in the workspace derives from that synthesized template: the Synthesized template tab shows it raw, and the Architecture tab is drawn from it — so the diagram can never disagree with your code.
What's free
A starter set of concepts and projects is free to read in full — no account needed — and one checkpoint is fully verified so you can feel the engine grade real code before paying. A membership unlocks verification, hints, and reference solutions across the whole catalog, in both languages.
From sandbox to your account
The sandbox proves your code synthesizes correctly; the last step of a concept takes it further — clone the finished template and run cdk deploy against your own AWS account, so you see a real URL answer. The deploy guide covers prerequisites, cdk bootstrap, and tearing down afterwards. Once you're done, remember to click Continue at the bottom of that lesson — like any reading lesson, it won't count toward the concept's progress until you do.