01
Architecture Brief
1 to 3 weeks
A compact system shape for a vague AI idea.
- Problem frame
- Workflow shape
- Risk map
AI-native system architecture
Make AI Ideas Buildable.
I turn vague AI and software ideas into working architectures, prototypes, and handover-ready specs.
Remote-first, async-friendly, with live calls used when they actually help the work.
Local-first systems
Evidence-bound workflows
Developer-ready handover
01 / What I do
I shape the system before production starts.
I work in the gap between a promising AI idea and a system a developer can actually build. The job is to define how information moves, what the model may do, where deterministic code takes over, and how the result stays inspectable.
Model agency is explicit
State changes are recoverable
Evidence is queryable
Specs survive handover
02 / What you get
Four useful outputs, depending on how far the idea needs to go.
02
Build Spec
3 to 8 weeks
A developer-ready plan with boundaries, data states, and build order.
- Interfaces
- Data ownership
- Implementation path
03
Reference Skeleton
1 to 4 months
A small working core that proves the system's main motion.
- Runnable core workflow
- Local state
- Testable path forward
04
Full Custom Prototype
3 to 8 months
A complete working prototype for unusual systems that need more than a plan.
- Custom architecture
- Working local system
- Handover-ready core
03 / Reference project
Reference project
The Ontology Machine
A local-first AI system for turning source material into evidence-bound corpus databases, queryable page evidence, semantic releases, and ontology lenses.
It is also the software artifact behind the position paper Making Human Suspicion Computable.
Source material
Evidence graph
Ontology lens
Inspectable output
What it proves
The Machine shows the kind of system I build: local-first, evidence-aware, agentic where useful, deterministic where needed, and documented enough to survive handover.
Public anchors
04 / Good fit
For ideas too complex for a prompt and too early for a build team.
I am useful when the hard part is not writing another model call, but finding the working shape of the system around it.
AI-native workflows with unclear architecture boundaries
Research ideas that need a developer-ready system shape
Prototype chaos that needs handover-ready structure
Local-first tools where evidence, state, and recovery matter
Not a fit: generic chatbot builds, polished SaaS delivery, or AI added where deterministic software is the better answer.
05 / Contact
Start with a short orientation call.
Send me a short email with what you have, what you want to build, what feels unclear, and what kind of outcome you are looking for. If it looks like a possible fit, we can schedule a roughly 30-minute call over a channel that works for both sides.
I generally work remote-first and async. Calls are used for orientation, clarification, and decisions, not as the main production surface.
The call is used to understand the project and decide whether a scoped offer makes sense. It does not start a project and does not create a paid engagement by itself.
If the project is a fit, I can send you a written offer that defines the scope, required material, exclusions, price, rough duration, delivery format, and acceptance conditions. Work starts only after the offer has been accepted in writing.
Scope
Required input
Exclusions
Price
Duration
Acceptance
Start by email