Cloud Platform Workbench
Turn a bounded request into a reviewable platform plan
A homelab proof of concept for checking infrastructure requests before they become implementation assumptions. Deterministic validation makes limits visible; a human approves a specific revision before the workbench creates its documentation pack.
Plan and review only · No infrastructure provisioning
Plan
documentation only
The workflow creates a planning pack. It does not provision infrastructure.
5
file pack
Architecture, ADR, delivery plan, operations guide, and manifest.
2
fixed patterns
VM with containers and one non-HA K3s server.
The problem
A request needs a visible decision boundary
A short environment request can leave important questions hidden: does the proposed pattern fit the sample limits, which assumptions are still open, and did the reviewer approve the current version?
The Workbench makes those checks and decisions explicit before producing a planning pack. Its examples are fixed synthetic designs, so the project demonstrates request handling and review flow rather than architecture generation or deployment readiness.
Workflow
Validate, review, then document
LangGraph coordinates the workflow state. Deterministic rules provide the sample checks; a human keeps the approval decision.
-
01
Describe the request
Choose one supported homelab pattern and provide a bounded set of environment requirements.
-
02
Validate the brief
Deterministic sample rules check individual and aggregate limits and show unresolved assumptions.
-
03
Review the plan
A person sees the proposed fixed design, caveats, capacity status, and the exact request revision before approval.
-
04
Generate the pack
Approval creates architecture, decision, delivery, and operations documents plus a manifest.
Demonstration scope
A synthetic review path
The PoC is designed to demonstrate four synthetic cases: rejecting a request that exceeds sample limits, clearing prior review after an edit, approving one fixed sample, and generating the five-file pack. The acceptance run and its results are not re-verified in this integration review.
Generated pack
Technology and role
A controlled workflow, not an infrastructure engine
The implementation combines Python, LangGraph, a request API and browser interface, with separate SQLite files for business decisions and workflow checkpoints. Fixed sample profiles keep architecture output predictable; versioned artifacts make the approval result inspectable.
Back to the portfolio
See how this planning PoC fits alongside the other selected work.