Internal UI / source system
User enters the order number or submits a request from an approved internal workflow.
Software leadership / AWS architecture
A concise AWS-native architecture answer for drafter-ready CAD automation: intake, object storage, orchestration, queueing, CAD runtime, validation reports, review flags, and operational observability.
Architecture mode
AWS-native control plane
Runtime caveat
CAD worker must be spiked early
Primary objective
Make automation reviewable, not magical
Target flow
The web request should start the job, not run the drawing generation synchronously.
User enters the order number or submits a request from an approved internal workflow.
Validate request, create a job record, resolve source package metadata, and return a job ID.
Store the DWG template, source drawings, PDFs, JSON, rules, SOW notes, and normalized manifest.
Coordinate gather, validate, generate, QA, package, and notify stages without blocking the web request.
Dedicated Windows, container, or APS route executes the CAD-specific drawing operation.
Return generated DWG, logs, validation report, review notes, and flagged assumptions.
Drafters see status, artifacts, confidence flags, and the areas that still need human judgment.
Operations can explain what ran, what changed, what failed, and which rule/template version was used.
AWS service map
The architecture stays simple: route the request, store packages, orchestrate stages, run the CAD worker, and make the result inspectable.
Web/API layer
Public or internal front door for submitting an order number, validating request shape, and returning a job ID.
Open routeRequest control
Lightweight request validation, job creation, metadata checks, notification hooks, and other short-lived control-plane work.
Open routeInput/output package
Object storage for DWGs, PDFs, JSON, templates, generated outputs, logs, validation reports, and reviewer artifacts.
Open routeOrchestration
Workflow orchestration when the automation has explicit stages, retries, branching, QA checks, packaging, or notifications.
Open routeBroker
Queue layer for decoupling the web/API request from long-running CAD generation and worker capacity.
Open routeJob state
Job records, status, input manifest, rule/template versions, review state, and durable audit metadata.
Open routeContainer worker
Containerized API or worker option when the CAD automation can run without installed desktop CAD or COM-style host behavior.
Open routeCAD runtime risk
Runtime option when full AutoCAD behavior, installed plugins, COM automation, AutoLISP execution, or licensed desktop constraints are required.
Open routeObservability
Logs, metrics, alarms, least-privilege permissions, encryption boundaries, and operational audit support.
Open routeCAD runtime decision
The first implementation spike should qualify licensing, desktop dependencies, DWG operations, plugin requirements, and whether full AutoCAD behavior is required.
Use when full AutoCAD, installed plugins, COM behavior, AutoLISP, or desktop licensing constraints control the outcome.
Use when generation can run as a containerized process without desktop CAD host assumptions.
Use when Design Automation covers the required DWG operations and licensing/runtime constraints are acceptable.
Use AWS for intake, storage, workflow, logging, and audit while isolating the CAD-specific runtime behind a worker boundary.
Validation and confidence
The goal is not to hide uncertainty. The output package should show what was deterministic, what was assumed, and what needs human review.
SDLC and observability
A useful first release improves the deterministic 60-70%, flags ambiguity, and does not pretend upstream data-quality issues are solved by automation.
Logs
job ID / order / input manifest / warnings
Versions
rule set / template / worker build
Reports
validation flags / reviewer notes
CI/CD
unit / integration / regression / smoke