I build software for work where a wrong answer is expensive.
Employment law, health benefits, federal data, and trucking. Five systems, each built on one rule: the software says what it can prove and hands the rest to a person.
Test counts from full suite runs on November 2, 2026.
Five systems, each with its problem, architecture, and results.
The same five rules in every build.
Rules before models
The parts that have to be right are deterministic code. A language model only writes the words around them.
Fail loudly
A failed check blocks the work. Nothing defaults to pass, and a broken data feed raises an alarm.
Show the source
Every number carries where it came from and when. If I can't cite it, the product doesn't say it.
Claims match code
Each product keeps a file of every promise and its build status. Anything unbuilt is labeled that way.
A person at the risky step
Software prepares the work. A human makes the call that carries legal or financial weight.
HR training, a builder's habits.
I'm Samuel Howell. I hold the SHRM-CP and a B.S. in Psychology from Oregon State University. Since 2018 I've run Great Green Yonder LLC, a forest-restoration contracting business that has delivered work under a USDA NRCS conservation contract.
In 2025 I started Foresight Labs (formerly Sage Studio) to build software for regulated work. I own the architecture, the specs, and the testing. AI coding agents help me write code faster, and nothing ships until it passes tests and compliance checks.
I'm looking for remote roles as a People Systems Architect or an AI Solutions Engineer.