ClinfoundPowered by Kinesthetic
Verified clinical coding · early warning · by API

Coding that never invents a code.

Clinfound reads clinical notes, proposes codes and early-warning flags, cites each one to the words that support it, and leaves every decision to a clinician.

Delivered by API. Runs on our infrastructure, you call it.
Early-warning results are in simulation, synthetic data only, not yet validated on real patient records.

Proven results · in our tests · measured, Archez internal, 2026

Zero invented codes.

The reader was scored against a 35B open LLM on the same clinical coding task. In our tests it invented no codes at all, scored higher, and ran on one CPU core.

Invented codes, same notes in our tests

0
Clinfound reader · codes not in the source
771
35B open LLM · codes not in the source
One mark for every code the LLM invented that was not in the source. The reader proposed none; a code that is not in the source is never proposed, and it says when it does not know.

Clinical coding F1 in our tests

0.557
Clinfound reader
vs
0.190
35B open LLM
Clinfound reader0.557
35B open LLM0.190
Same task, same notes, same scoring, in our tests.

Speed in our tests

~5,000x faster
than the 35B open LLM on one CPU core. Median ~44 ms per note.
Clinfound reader
~44 ms median per note
35B open LLM
The lower bar is cut; it is not drawn to scale.
CPU-only and cheap, built to run over a whole population rather than a sample.
Illustration · how a question becomes a cited answer

Watch a note become a traceable proposal.

A scripted walk-through of what you get back. Hover any proposal to see the exact words it is cited to.

Illustrative · scripted · not a live engine call
Example note (fictional, no patient data)
What Clinfound returns
ReadsFinds supportProposesCitesWaits for clinician
Terms shown as words; code identifiers are left out of this illustration.

The illustration is a fixed script that shows the shape of a result: a proposal, the words it is cited to, and a wait for a clinician. It is not a live call to the engine and does not show real output.

Early warning · population health · by API

Cancers flagged months before diagnosis.

Symptomsrecorded in notes
Flaggedby Clinfound
median 3.2 months earlier
Diagnosisas recorded
44%
of cancers flagged by their own symptoms, a median 3.2 months before diagnosis.

Clinfound applies NICE NG12 suspected-cancer guidance as rules across the notes you already hold, and proposes a flag for a clinician to review. It never refers, orders, or acts.

In simulation, synthetic data only, not yet validated on real patient records. The 44% and 3.2-months figures come from a simulation on synthetic data. They are not a clinical result and should not be read as one.

A gift to the world, delivered by API

Early warning should not depend on who can afford a GPU estate. Clinfound runs CPU-only on our infrastructure. A health system calls the API over the records it already holds, and gets back flags that a clinician can check to source and act on, or set aside.

Validation on real patient records is the next step and is not yet done.

What the engine is, from the outside

Powerful but inert.

Clinfound is Powered by Kinesthetic. These are its properties as you see them from the outside. Hover or tap any one for the rest.

Verified

Every proposal is checked against its source before it leaves.

Nothing is served on a guess.

Traceable

Each code or flag points to the exact words that support it.

A regulator, hospital or court can audit every answer to source.

Deterministic

The same note gives the same answer every time.

So results can be reproduced and audited.

No hallucination

In our tests it invented no codes at all.

It cannot make up a source that does not exist.

Proposes, never acts

It cannot refer, order, record or act.

A human decides and does.

Every answer cited

No bare answers.

Each one is a readable, cited step you can check.

Says when it does not know

If the note does not support a code or a flag, it says so.

It does not fill the gap.

CPU-only, cheap, fast

Median ~44 ms per note on one CPU core in our tests.

No GPU estate needed to run it over a whole population.

Cannot be argued across its lines

Its rules are hard lines, not suggestions.

It cannot be argued or jailbroken across them.

Safety

Safe by construction, not safe by supervision.

Kinesthetic has no wants, no needs, no awareness. It does not think. It works only from verified rules and traces every step.

It cannot act in the world. It only ever proposes; a human decides and does. Its rules are hard lines, not suggestions, so it cannot be argued or jailbroken across them.

Every output is a readable, cited step you can check. A regulator, a hospital or a court can audit every answer to source.

Clinfound is a product name, not a clinician. It supports coding and case-finding decisions; it does not make them.

How you use it

You call it. We run it. A clinician decides.

Send notes to the API

Clinical coding and early-warning case-finding are both delivered by API. There is nothing to install and no model to host.

It runs on our infrastructure

CPU-only, deterministic, and fast. Served answers leave; the engine stays with us.

You get cited proposals

Each code or flag arrives with the words it is cited to. A clinician accepts, amends or sets it aside. Clinfound never acts.

Request access

Clinfound is offered to health systems, coding teams and research partners by API. Tell us what you want to read and we will come back to you.

Placeholder contact address while the public site is being prepared.

Request access