From a paragraph to a buildable architecture
Eight stages, each one producing something you can inspect before the next begins. No stage decides anything on your behalf without labelling it as a decision.
Step 1
Describe what you are building
A paragraph is enough. The engine reads it for the application type, the regions involved, the rough scale and whether AI or sensitive data is in play, and starts with that.
Produces A starting profile, with each reading marked as stated or inferred
Step 2
Answer the interview
Questions are ordered by how much they change the answer, and they adapt to what you have already said. The catalog covers fifteen domains including data, geography, compliance, reliability, team and budget.
Produces Confirmed requirements, with assumptions labelled as assumptions
Step 3
Review what was captured
Every requirement shows its status - confirmed, inferred, assumed or unknown - so nothing unverified quietly becomes a fact. This is the last cheap moment to correct something.
Produces A requirements set you have actually looked at
Step 4
Compare the options
Candidate architectures are matched against your confirmed requirements and costed. Cost, operational effort, lock-in, risk and compliance readiness are compared side by side.
Produces A costed comparison, not a single recommendation
Step 5
Read the economics
Each option carries a cost model built from the architecture it describes, with estimates clearly separated from anything confirmed. GPU and API break-even is calculated rather than estimated.
Produces Monthly cost models you can argue with
Step 6
Check the risks and the compliance story
Risks are listed against the option they belong to, with the technical controls that mitigate them. Compliance output maps controls to framework considerations and says what falls outside architecture scope.
Produces A risk register and a compliance readiness view
Step 7
Select and record the decision
Choosing an option writes a decision record: what was decided, which requirements drove it, and what was rejected. That is what makes the reasoning reviewable six months later.
Produces Decision records and an implementation plan
Step 8
Take the artifacts
Generated Terraform, the implementation plan in order, and a shareable report. Nothing is applied to your infrastructure: you review the code and deploy it with your own process.
Produces Terraform and a report you can circulate
What that workflow deliberately leaves out
- It never connects to your cloud accounts, so nothing in the sequence can change your production infrastructure.
- It does not apply the Terraform it generates. Validation of that code is scheduled, not built, and generated files say so.
- It does not produce a compliance certification, and it does not guess a value you did not give it - gaps stay visible as gaps.
Try the first stage now
The interview teaser on the home page reads a description and shows you the questions it would ask. It needs no account.