Overview

The Summarization API writes a summary of clinical documents, following your instructions. Send a document’s text — or several documents about one patient — and a prompt saying what the summary is for: who will read it, what to cover, how long it should be. You get back one summary, as Markdown.

The summary takes negation (“no evidence of metastasis”), uncertainty and history into account. With several documents, the summary cites them, so your users can see where each statement came from.

How a request works

The API is asynchronous. You submit a job, wait for it to finish, then download the result once.

POST   /v1/jobs              → 202, the job: {"id": "job_…", "status": "queued", …}
GET    /v1/jobs/{id}         → status: queued → running → succeeded | failed | canceled
GET    /v1/jobs/{id}/result  → the result, once; a second request returns 410
DELETE /v1/jobs/{id}         → cancel the job, or delete its unread result

A single document usually takes a few seconds; a dozen documents take about half a minute. Poll the status every two seconds or so.

What you can send

The last four can be combined in one job. However the documents arrive, you get one summary.

How well the summary serves its reader depends mostly on the prompt: see Writing prompts.

Versioning

The API is versioned in the path (/v1). Changes within v1 are additive: new optional fields, new error codes, new kinds of input. Ignore fields you do not recognise. A breaking change will come as a new version, with advance notice.

Every result carries a pipeline_version, which identifies the release that produced it.

Support

Email support@emtelligent.com, quoting the request_id from an error response or the job id.