Human v01
The skeleton we start every build from.
A deliberately small architecture for API-based services that have to carry AI without being rebuilt around it.
the skeleton
Bounded contexts, one public surface.
Human v01 is deliberately small. Every app is a bounded context with a stable mount point and every response shape is a serializer contract, so you grow the system by adding slices rather than by rewriting the skeleton.
What is in it
- Python and Django, with a headless CMS for content
- One public API surface, versioned, with a published schema
- Token identity and hashed keys for machine callers
- Background work on a queue, not in the request
- A lint and test gate that runs the same way everywhere
What it is for
- Payments, messaging and comms as ordinary integrations
- A stable mount point per service, so one can be replaced
- Environment differences as overlays, never as forks
- A frontend that talks to the same API everyone else does
why it suits AI
Microservices are what make a model replaceable.
AI moves faster than the systems it sits in. The architecture has to assume the model will change and the product will not.
A model is a dependency, not the design
Behind a stable internal contract, swapping a provider is a deployment rather than a rebuild.
Integrations are the product
Payments, messaging, notifications and scheduling are where the value lands. The AI is one more service beside them, not a layer on top of everything.
The gate does not move
Same tests, same lint, same deploy, whether the change is a migration or a new prompt. That is what makes shipping AI work feel ordinary.