About
Built by someone who
works on medical equipment
Shellfinity is the work of Daniel Bearden, a biomedical engineer at Kaiser Permanente, an autodidact, and a father of two with a third on the way. He built FROS because a system that reasons through a long, high-stakes problem should stay consistent with itself, and the people who rely on it should be able to see why it landed where it did.
The founder
Daniel Bearden
Biomedical engineer by trade. Self-taught in formal methods, proof assistants, and language modeling. Husband and father of two, with a third on the way.
Years of working on clinical equipment shaped one conviction that sits at the center of this project: when the answer matters, the reasoning has to be visible. A device earns its place in a hospital, a courtroom, or a compliance workflow by explaining its own output. A language model is held to the same bar.
Shellfinity exists to make that standard practical. Every ruling from the engine is reproducible, inspectable, and anchored in a mechanically checked proof.
The work
What FROS actually is
A record that holds steady
As a model works, the facts and constraints it commits to are held in a verified record outside the model. The part that checks new steps against that record is a deterministic function with zero learned parameters. Same state, same step, same answer, every time, however long the task has been running.
Proven on the hard cases first
The clearest way to test a checker is to point it at domains where a wrong call has a cost. Resolving what a word means in context sits at 94.5% on the standard benchmark. Patient findings are checked against 30,981 diagnoses, each exclusion carrying a written reason. These prove out one capability, and the product is larger than any one of them.
A set of constraints that grows
When a new fact enters the record, the engine works out what it implies and keeps the rest consistent on its own. The constraint set can keep expanding while the engine, rather than a person, re-encodes each case, which is what turns a checker into something a model can lean on for the long haul.
Why this exists
Reasoning you can hand to a professional
In the domains that matter most, clinicians, lawyers, compliance officers, auditors, a confidence score is worthless. They need to see the reasoning and be able to disagree with it in specific terms. That is what Shellfinity ships: a verdict a professional can read and a record a professional can challenge, in place of a black box with a percentage on top.