The Omnivorous Health Data Ingestion Engine®
A photographed lab report. A sentence said out loud. A video diary. A PDF from a portal. A handwritten page. None of it is usable until something reads it. Charybdis turns all of it into one canonical record where every fact carries where it came from, how it was read, when it happened, and how much to trust it.
The problem
Almost every organization in healthcare has the same complaint and describes it differently. Providers say the record is incomplete. Payers say the data arrives too late to matter. Employers say engagement is low. Digital health companies say onboarding takes too long. Underneath all four is one problem: a person's health history exists — it is simply not in a form anything can use.
The forms a person's health history really takes
One record, one shape, whatever the origin
The usual answer is a person retyping numbers or importing records. That works, at a cost, and introduces a quieter problem: a retyped number carries no record of where it came from. Six months later nobody can say whether a value came from a certified laboratory or from someone's memory of what the nurse said.
What it does
Charybdis is the layer that sits underneath a health product and turns whatever a person has into one consistent record. The output is deliberately boring, and that is the point: a blood pressure read from a photograph and one arriving from a hospital system look identical to whatever consumes them — except in the provenance, where the difference belongs and can be inspected.
PDF exports and printed reports, read from the text layer where there is one.
A phone photograph of a report, or a page from a kitchen drawer.
A spoken sentence becomes a dated, coded medication event.
What was said and what was shown, both read and timestamped.
FHIR resources and bundles, already coded at the source.
Wearables, pharmacy fills and event subscriptions.
Provenance is the product
A person has the information. No system does. Nothing here can be charted, compared with last year, or acted on without someone retyping it.
What makes it different
Each one is a commercial position as much as a technical one.
Every value keeps the document, the page and — for video — the second it came from. A number that cannot be traced is a number that cannot be defended, and in a regulated setting that is worse than having no number at all.
A reading outside physiological possibility, or read from a blurred image, is flagged and kept rather than quietly dropped. Suppressing a doubtful reading hides a decision from the person it belongs to.
When two sources report different values for the same thing, both are preserved and the conflict is surfaced. When they agree, that agreement is recorded as corroboration rather than deduplicated away.
Speech recognition, image reading and clinical terminology are replaceable components behind a stable boundary. A customer with an existing vendor contract can bring it. No supplier can hold the product hostage.
Video
A patient records a short diary entry — "my sugar was one-forty this morning" — while holding the meter up to the camera. Transcribe that and you keep half of it.
The spoken number and the number on the screen are two independent pieces of evidence for the same fact, and when they agree that agreement is worth recording. Both are read, and every resulting value is stamped with the moment it came from.
The easiest thing a patient can do is talk into a phone. Making that a usable input removes most of the friction from the thing you actually want them to do.
hemoglobin 14.2 g/dL LOINC 718-7 frame 0 · 0.0s · conf 0.99 platelets 244 K/uL LOINC 777-3 frame 1 · 4.0s · conf 0.99 fasting_glucose 140 mg/dL LOINC 2345-7 frame 2 · 8.0s · conf 0.97
Every fact stamped with the second it came from, so a reviewer can play back the moment a value was said or shown rather than taking it on trust.
Where it sits
Charybdis answers one question: what information is contained in the available data? It does not score, rank, advise or diagnose. Those are the next questions, and they belong to whatever product sits above it.
What is in the data?
Ingestion, extraction, normalization, provenance and reconciliation — the work that has to happen before anything else can.
No opinion attached
No score, no risk tier, no recommendation. Normalized evidence, and the record of where every piece of it came from.
Consumer-neutral by design
The engine has no knowledge of which product reads its output, which is precisely what makes it licensable on its own terms.
That separation is deliberate. A health system, a payer analytics group or a digital health company that wants traceable evidence and no opinion attached is a legitimate customer, and keeping the engine indifferent to its consumer is a commercial decision rather than an accident of architecture.
Charybdis is available for licensing as a standalone ingestion layer, separately from the products built on it. If you are assembling health data from more than one kind of source, we should talk.