Compile
A guideline chapter or an institutional protocol goes in. AI proposes the rules, a clinician reviews and signs off, and the pack is versioned.
Xondis
Every guideline is public.
Almost none of them run.
Xondis compiles published clinical guidelines and institutional protocols into versioned, deterministic software. The same patient produces the same recommendation every time, and every recommendation traces back to the line it came from.
A document cannot execute.
The standard of care is not a secret. It is published every year, in detail, by the societies that define it. Anyone can download it.
And then every hospital, every medical group, every value-based care organisation takes that same public document and reimplements it by hand. Someone reads the chapter. Someone else builds the order set. A third person writes the registry query. A fourth checks, months later, whether any of it matches what the guideline actually said.
Then the guideline is revised, and all of it happens again.
The tax code is public too. Nobody reads it. They open software that executes it, files the return, and shows its work. Clinical medicine has the published rules and none of the software that runs them.
AI at compile time.
Determinism at runtime.
Xondis reads a guideline or a protocol and compiles it into a knowledge pack: an executable set of rules, versioned like software, with every rule bound to the source text that produced it.
That compilation step is where the language model works. It is slow, reviewable, and done once. A clinician reviews the output before anything is published.
What runs in production is not a language model. It is the compiled pack. Given a patient state, it returns a recommendation, the same one, every time, with a citation attached.
- 01Guideline or protocol
- 02Compiled knowledge pack
- 03Deterministic runtime
- 04Recommendation with source line
Execute
The pack runs against patient data and returns a recommendation. No model call, no drift, no temperature.
Audit
Every recommendation carries the rule that fired and the source line behind it. Every version is diffable, so when the guideline changes you can see exactly what changed.
You can check our work.
Determinism is not a marketing word here. It is a testable property, and these are the tests.
Same input, same output
Run the same patient through the same pack a thousand times and you get one answer, not a distribution. The output has a digest. It either matches or it does not.
Order does not matter
Shuffle the sequence in which rules are evaluated and the result is unchanged. A pack that produces a different answer depending on evaluation order is not a pack, it is a bug.
Every recommendation has a provenance line
Not "consistent with guidelines". The specific sentence, in the specific version, that produced this specific recommendation. A clinician can read the basis and disagree with it.
Versions are diffable
When a society revises a chapter, you get a diff of the compiled pack, not a memo telling you to re-read the PDF.
Three sources, and one of them is already yours.
Published guidelines are copyrighted. That is a real constraint and we would rather state it than let you discover it.
If you are measured on it,
it has to execute.
A protocol nobody follows is not a clinical problem in the abstract. It is a number somebody is scored on.
Kidney health evaluation for patients with diabetes. Optimal dialysis starts. Sepsis bundle compliance. Star ratings, HEDIS, quality measures tied to shared savings. Each of those is a published rule that somebody reimplemented by hand, and each of them has a person whose year depends on the result.
Xondis is for that person.
It starts without patients.
The first step is a concordance test, and it needs nothing from your clinical environment. No integration, no patient data, no commitment.
We compile one protocol you already own. We run it against your own document and show you, rule by rule, where the compiled pack agrees with what you wrote and where it does not. The disagreements are the interesting part, because they are usually places where the written protocol was ambiguous and everyone had been resolving it differently.
If that is useful, there is a next step. If it is not, you have lost a conversation and gained an audit of your own protocol.
Built by a physician who got tired of the gap.
Xondis is a product of Endo Mind, Inc., founded by Jean Jacques Nya Ngatchou, MD, a board-certified endocrinologist who left full-time practice to build it.
Bring us a protocol.
One protocol you already own, and a measure you are graded on. That is enough to start.
Contact XondisDo not include patient information.