Blueprints
How I'd architect controlled AI systems
These are the reference architectures I sketch out when I'm thinking through how AI should work inside regulated environments — retrieval governance, supervised agent workflows, evaluation, and evidence-oriented design. Concepts, not products: synthetic data, human review built in, nothing autonomous.
Reference Blueprints
Choose an educational concept
Select a blueprint to explore educational architecture, data flow, and controlled usage patterns. These are not products for sale or implementation offers.
Regulatory Intelligence Blueprint
Public regulation text to reviewable obligation candidates
An educational reference architecture for turning public regulatory text into structured obligation candidates, control-mapping drafts, and evidence-pack templates. Outputs require human validation and do not constitute legal or regulatory advice.
2,847
Obligation Candidates
2,401
Draft Mappings
127
Review Gaps
94%
Template Coverage
Sources
Review Window
Synthetic timeline
Reference Architecture
flowchart LR
A["Public regulatory text"] --> B["Clause and obligation candidate parser"]
B --> C["Control mapping draft"]
C --> D["Evidence-pack templates"]
D --> E["Internal reviewer workflow"]
E --> F["Human validated output"]