01
Find the right boundary
Determine where AI contributes judgment—and where deterministic software should take over.
/ Applied AI Systems
Complexsystems, madeclear enough to use.
I design, test, and troubleshoot AI-enabled workflows—then turn the evidence into decisions people can understand.
Selected work / 03
The work starts where the assumption breaks.
Three cases across software, governance, and a real operating problem. Read the finding first; inspect the technical evidence when you want it.
AB. / VIS-TAG-0101
When AI Is the Wrong Tool
Finding
A model was being asked to perform a transformation ordinary software could guarantee.
What changed
The unreliable AI step became a deterministic, zero-token lane that produces the same answer every time.
Normalization · validation · local routing
AB. / VIS-GOV-0102
Building a Governed Local AI Application
Finding
Useful model output was being confused with permission, memory, and system authority.
What changed
The application now keeps model proposals, software enforcement, and human decisions separate.
Local application · correction history · synthetic fixture
AB. / VIS-STUDIO-0103
Refusing to Design Before Measuring
Finding
A polished technical plan can still be wrong when the underlying room evidence is incomplete.
What changed
A readiness gate now stops fabrication advice until the required facts are present and attributable.
Business intervention · evidence capture · deterministic gate
Practice / Where I help
I work between the technical system and the people who need to trust it.
02
Make failure inspectable
Design evaluations that preserve what failed, what changed, and what remains uncertain.
03
Build governed workflows
Separate model output, source material, system rules, and human authority.
04
Translate the evidence
Turn technical behavior into clear decisions, documentation, and client communication.
AB. / Flagship system 01 / AlexOS
Follow the question further.
AlexOS is a governed environment for discovering, testing, preserving, and teaching understanding. This is where the method, Workshop, Trials, and deeper system evidence live.
Currently / In development
The work remains open where the evidence remains open.
01
Visit the Open Lab Evidence incomplete
Router trace explorer
A sanitized end-to-end trace and ablation record remains in development before the interactive explorer can be presented as earned.
02Reproduction open
What a system can know about itself
A mixed recursive result is preserved while fresh-session, failure-injection, trace, and ablation tests remain open.
03Active protocol
Financial research workbench
The evidence and governance contract is being tested before any data provider or operating system is selected.
A real question / A useful start
Bring the problem before the solution.
If you are evaluating an AI workflow, untangling a system failure, or deciding whether AI belongs in the work at all, send a short description of the problem.
Do not send confidential, regulated, privileged, or security-sensitive material in an initial email.