An ambient scribe API is a service through which the makers of patient records build an AI medical scribe into their own product, instead of building one themselves. Here, an API is a set of instructions that one program can send to another over the internet. The record maker's software sends the recording of a visit to the scribe company's service, and receives a draft note in return. Products like this are sold mainly to record makers whose customers are in the United States.
Why would a record maker not build its own scribe? Building one needs speech recognition that understands medical words, a language model that writes good notes, and years of testing with doctors. A company that makes records for physiotherapists, or for eye clinics, cannot easily build all of that. But its customers now expect a scribe. So it rents one from a specialist, and shows the scribe inside its own screens, under its own name.
Consider a company that makes the record used by 3,000 physiotherapy clinics in the United States. Its engineers add a Record button to their app. During a 30-minute session, the app sends the audio to the scribe service as the session goes on. When the session ends, the service sends back a draft note within about a minute. The note is not one block of text. It is divided into the sections that the physiotherapy record uses: the patient's pain score, the exercises done, how far the knee bends, and the plan for the next session. The app places each section in its own field. The physiotherapist checks the note and signs it, as with any AI medical scribe.
The hard part is fitting the note to the record's own fields. A general scribe writes a note for a family doctor. A physiotherapy note needs different sections and different words. So the service lets each record maker define its own note template, and the language model fills it in. Picking the right facts out of the conversation for each field is a task of clinical NLP.
The record maker usually pays the scribe company a fee for each doctor or therapist who uses the scribe each month. It may charge its own customers extra for the feature, or include it in its price.
Many software companies face the same choice in other areas: build a feature themselves, or add a specialist's feature to their own product. In payments, adding a specialist's payment service to a product is called embedded payments. Behind the scribe service are models for speech and language, described in speech and voice AI.