AI patient intake structuring.
The OneDose Intake Agent turns messy inputs into structured records. Handwritten prescriptions, faxes, phone photographs and PDFs go in; drug, strength, dose, frequency, duration and prescriber come out, mapped to the fields the pharmacy system expects. Anything the agent cannot read with confidence is flagged for a human rather than guessed.
Intake is where the rest of the medication journey gets its data. A wrong strength captured here is a wrong strength in every downstream reminder, refill and check-in.
How does the Intake Agent actually work?
What follows is the mechanism — what the agent does, in what order, and where a human stays in the loop.
It reads what the pharmacy actually receives
Not a clean HL7 feed — a photograph of a handwritten script taken at an angle in bad light, a fax that has been through a fax twice, a PDF export with the table structure gone. That is the real input, and it is the input the agent is built for.
It maps to a structured record
Drug, strength, form, dose, frequency, duration, quantity and prescriber, in the fields the downstream system expects — so the output is a record rather than a transcription.
It flags rather than guesses
The important behaviour is what happens at low confidence. An illegible strength is escalated for human review, not resolved to the most probable value — because the most probable value is wrong often enough to matter when the field is a milligram count.
It shows its working
Each structured field is traceable back to the region of the source document it came from, so a reviewer checks a flag in seconds rather than re-reading the whole script.
What will the Intake Agent not do?
The limits matter more than the capabilities when the subject is medication, so they are stated plainly rather than left to be discovered in a pilot.
- It does not clinically review a prescription. Structuring what the script says is a different job from judging whether the script is right.
- It does not resolve an ambiguous field by picking the likeliest value. Low confidence escalates.
- Accuracy depends on the source. A legible e-script and a photograph of smudged handwriting are not the same problem, and no vendor number covers both honestly.
How does this agent fit with the others?
The Intake Agent is one of six OneDose agents that share one patient context and one audit log. That shared context is the difference between a workforce and six tools: the voice agent already knows what the follow-up agent asked yesterday, so the patient is not asked to repeat themselves and the care team reads one history rather than six.
Frequently asked
- What is AI patient intake structuring?
- AI patient intake structuring is the conversion of unstructured clinical inputs — handwritten prescriptions, faxes, photographs, PDFs — into structured, machine-readable records with the drug, strength, dose, frequency and prescriber in defined fields, so downstream systems can act on them without a human retyping the document.
- What happens when the agent cannot read a prescription?
- It flags the specific field it could not read and routes the document for human review, with the uncertain region highlighted against the source image. It does not fill the field with its best guess — on a strength or a frequency, a confident wrong answer is worse than an admitted unknown.
- Does the intake agent check whether a prescription is safe?
- No. The Intake Agent structures what the prescription says. Clinical review of whether the prescription is appropriate is a pharmacist’s job, and structuring the record is what gives the pharmacist something to review.