Every record on file, and where each value came from

A veterinary record arrives as paper, a PDF or a photograph, and it is only useful once the values inside it are on a chart.

Talli does that reading, and it shows its working: every clinical fact carries provenance back to the document it came from, so you can always ask where a number came from.

Putting a document in

Upload a file, choose a photo, or photograph a page. The file is checked before it is read: it is streamed to a malware scanner, and its real type is confirmed from its own bytes rather than from its name.

A PDF may be up to 50 MB, an image up to 25 MB, a DICOM file up to 200 MB and a text file up to 10 MB, and each ceiling is enforced twice, on what the client declares and again on what actually arrived.

50 megabytes
megabytes, a PDF
25 megabytes
megabytes, an image
200 megabytes
megabytes, a DICOM file
10 megabytes
megabytes, a text file

What happens next

The document moves through ten named stages: received, stored, classified, matched to a pet, extracted, validated, confidence assessed, applied to the record, indexed and notified. At every one of them the pipeline may stop and ask instead of guessing.

The queue where a person decides

Anything the pipeline will not file on its own goes to a review queue, and a person resolves it. There are seven reasons a document waits there, and each one is a specific question rather than a generic error.

The date. The page does not say what day it is from. Nothing dated by Talli instead of by the document is filed.
The kind. It is not clear what kind of document this is, and the fields read out of it mean nothing until that is settled.
Low confidence. What was read is shown next to the document so you can check it before it is filed.
An implausible reading. A value the animal could not have produced is held, and the bound it failed is named.
A duplicate. The bytes are identical to a document already on file, so nothing from the copy was added.
A contradiction. The document disagrees with the record. Nothing is changed, and the queue does not choose between them for you.
The pet. The document could not be matched to one animal with enough confidence, so it stops rather than filing against a guess.

Thirteen kinds, and one that files nothing

A document can be filed as a visit summary, laboratory report, prescription, vaccination certificate, invoice, imaging report, imaging study, discharge instructions, referral, insurance, diet plan, consent form, or "something else". The last one is offered with its consequence in the label, because "none of these" is a true answer and nothing is filed from it.

A document nobody has classified yet reads "not yet classified", which is the absence of a type rather than a type.

What is never done

No value is dated by Talli. If the document does not carry a date, the values wait for one.
No half-read document is filed. A row that fails validation is not written, and there is no "apply the good half" path.
Nothing read by optical character recognition is applied without a person looking, at any confidence at all.