I'm building Lucy AIOS, an experimental local-first AI operating architecture focused on a question I think matters more as AI systems become increasingly capable:
How do we increase intelligence without silently increasing authority?
My work centers on separating reasoning, authority, execution, verification, and evidence so AI models and specialist agents can contribute useful work without becoming independent roots of trust.
Lucy is the operating intelligence and owner-facing system.
The architecture is built around a simple lifecycle:
Observe → Understand → Propose → Govern → Execute → Verify → Record
Lucy is designed as one coherent operating intelligence that can work with multiple specialist models and environments while maintaining shared task identity, context, governance, and evidence.
- Local-first AI
- Human-governed execution
- Typed capability boundaries
- Verifiable actions
- Durable evidence
- Context and task identity
- Specialist-model orchestration
- Sandboxed software engineering
- Security observation
- Digital-twin and world-state research
- Planetary-system modeling
E.M.M.A. — Enhanced Machine Mind Architecture
E.M.M.A. is the governance and evidence layer behind Lucy.
Intelligence can reason and propose.
Authority is separate.
The goal is to make important AI activity inspectable:
- What was requested?
- What understood the request?
- What capability was selected?
- What authority existed?
- What actually executed?
- Was the result verified?
- What evidence remains afterward?
That distinction between proposal and permission is one of the foundations of the architecture.
Helix is the governed engineering environment being developed around Lucy.
Specialists can inspect architecture, propose implementation work, write code, test changes, and critique one another — but engineering work ultimately returns through Lucy's governance path.
The intended model is not:
AI agent → machine
It is closer to:
Human → Lucy → specialist → proposal → governance → sandbox → verification → evidence
The architecture is also being tested through real application environments.
A construction and insurance-claim project platform focused on:
- field evidence
- project workflows
- provenance
- estimates
- documentation
- tenant isolation
- human-gated claim readiness
- governed AI participation
Ongoing work includes:
- Earth observation ingestion
- planetary boundaries
- hazard modeling
- digital twins
- Twin Earth concepts
- NVIDIA Earth-2 / OpenUSD foundations
- prediction provenance
- cascading-system analysis
These areas are gradually converging into a broader planetary ecosystem and hazard-response architecture.
The primary public technical entry point is:
Lucy AIOS — Public Architecture
It contains public architecture, design principles, current-status material, verification evidence, and selected proof-of-concept documentation.
The active development system is substantially larger and remains private.
A few rules remain consistent across the work:
One Lucy. Many specialists.
Memory can affect context. It cannot grant authority.
A model recommendation is not an execution permission.
Designed ≠ built ≠ wired ≠ verified ≠ live-proven ≠ production-ready.
Important actions should leave evidence.
I'm a founder and independent systems architect developing these systems with AI-assisted engineering tools and specialist models.
My background didn't begin in traditional computer science. A lot of this work grew out of asking practical questions about how increasingly capable AI systems could remain understandable, governable, and useful to the person operating them.
The repositories here document that evolution — including experiments that worked, architectures that changed, and systems that are still being built.
I'm especially interested in connecting with people working on:
AI architecture · agent governance · local AI · AI safety engineering · digital twins · simulation · planetary systems · construction technology · applied AI
Active research and development — 2026