Fit for discharge by ten. Between the decision and the door, a day of writing.
The encounter is documented as it happens. The hours either side of it did not move: the discharge summary assembled by hand, the medication list reconciled across four sources, the brief a consult team reads before it can start. Exemplary AI drafts them from the indexed stay, cites the page behind every line, and nothing moves until a person signs.
The note is written. The summary is not.
The stay is documented as it happens — an admission note, daily notes, an imaging report, two medication lists. Turning that record into the documents a discharge needs is still done by hand, by the people a ward can least spare. None of it is a knowledge gap.
The hospitalist writes the stay twice.
Everything the discharge summary needs already exists in the daily notes. Assembling it is the evening's work, after this round and before the next one — and the patient left hours ago.
The pharmacist reconciles four lists.
The list taken at admission, the inpatient chart, the outside letter, and what the patient says they take. Line by line, against the clock, before anyone can safely leave.
The consult starts with a hunt.
The question that prompted it is one line in yesterday's note. The specialist reads backwards through the stay to find it before the visit can begin.
Between the decision and the door.
One stay, five source documents, one draft, one flag, one signature. The whole interval, drawn — including the row the agent refuses to decide.
The decision is made
On the morning round: fit for discharge. The clinical work is done. Everything between this mark and the door is writing — and the stay is the longest wait a bed has.
Drafted from the record. Signed by a person.
Nothing here listens to a conversation, and nothing here decides. Everything below reads what the stay already wrote down, drafts the document the next step needs, and stops at the signature line.
- Course of the stay, from the daily notesp.2
- Imaging, quoted from the reportp.9
- Follow-up and the letter homep.11
Unsigned — nothing has filed
Assembled from the stay, cited to the page.
Drafted from the indexed record, not from recall, while the team is still on the ward. A person reads sentences and checks them — every line carries the page it was read from.
- Drafted from the indexed stay, not dictated from memory
- Every line carries the page behind it
- The medication section arrives flagged for reconciliation
- The hospitalist signs before anything files
Fourth row — for the pharmacist
Compared across sources. Flagged, never resolved.
The documented lists are set side by side — admission, chart, outside letter. Where they agree, the rows tie. Where they disagree, the agent raises a flag for the pharmacist. It does not choose a line, and it does not prescribe. That is a person’s decision, kept that way on purpose.
- Every documented source, one comparison
- Agreement shown as ties, disagreement as flags
- The pharmacist decides; the agent never does
“The consult question, quoted from the requesting note”
- The relevant days of the stay
- The most recent imaging report
- The current medication list
The brief is ready before the specialist walks in.
The reason for the consult, the relevant stretch of the stay, and the most recent imaging report — one page, assembled from the indexed record, each line pointing at its source. The visit starts at the question instead of the chart.
- The consult question, found and quoted
- The relevant days of the stay, not all of them
- Each line points at the page it came from
Replies drafted in the reader's language.
Discharge instructions, the letter to the practice, the follow-up reminder — drafted from the signed summary, in the language the reader speaks. 120+ supported, and a person still signs before anything goes out.
- Drafted from the signed record, never from scratch
- 120+ languages, the same content in each
- Nothing sends without a signature
The round decides at 11:05. Here is everything that happens next.
No step skipped, and no step automated that shouldn't be. Three of the six belong to people, and the platform keeps it that way.
Draft
- Admitted from the emergency department; the course of the stay is summarised from the daily notes. p.2
- The most recent imaging is reported in the record, and the report is quoted rather than recalled. p.9
- Follow-up is arranged; the letter to the practice is drafted in the language its reader speaks. p.11
Medication list — two sources compared
- First row — documented in both lists
- Second row — documented in both lists
- Third row — documented in both lists
- Fourth row — documented differently in the two lists
Flagged for the pharmacist. Not resolved by the agent.
Nothing leaves the chart until this row is signed.
- The round
Fit for discharge.
The decision is clinical and it is made by clinicians, on the round. The platform is not in the room. Its work starts the moment the decision is in the note.
- The record
The stay is already indexed.
Admission note, daily notes, imaging report, medication lists — indexed as they were written, visible only to the care team that owns them.
- The draft
The summary is drafted, with citations.
One document, assembled from the stay. Every line carries the page it was read from, and the line the record cannot support is shown as missing rather than guessed.
- The pharmacist
One row is flagged, and a person resolves it.
The medication comparison ties three rows and flags the fourth. The pharmacist decides which line stands. The agent recorded the disagreement; it does not get a vote.
- The hospitalist
The draft is read, corrected, signed.
Verification is opening the citations, not re-reading the stay. The signature is the first moment anything is allowed to leave the chart.
- The door
Filed, sent, and logged.
The summary files through the EHR connector. The letter to the practice goes out in the reader's language. The audit trail records who approved what, and when — and the bed is a bed again.
The paperwork stops queueing behind the people who can sign it.
The writing that used to wait for the evening is drafted while the team is still on the ward, and the people the ward can least spare go back to being checkers and signers instead of typists. The interval between the decision and the door stops being the longest part of the stay.
A consult begins with the brief, because the record that drafted the discharge summary answers the specialist’s first question too.
A difference between two medication lists is a flag on a screen before the patient leaves, not a phone call to the ward after.
The stay was indexed once, so the next document — the letter, the reminder, the reply — costs a configuration, not a project.
The chart never leaves the building. Neither does the draft.
A hospital’s vendor review is a list of places patient data goes. Honest framing: most AI adds a line to that list. This deployment removes one — the models run inside your network, under your BAA, on your hardware.
Inference where the chart lives
Bare-metal, private cloud, or fully air-gapped. The stay, the draft and the flag never cross your perimeter — not even for inference.
Scoped to the care team
Groups and IAM, least privilege, deny overrides allow. A ward sees its own patients; a service sees its own consults; nobody sees more because a model is involved.
One fewer sub-processor
AES-256 at rest, TLS 1.3 in transit, a BAA-ready architecture — and no third party in the inference path for your privacy office to assess.
Deployed inside your perimeter under controls that already exist. Model updates arrive by secure transfer, so an air-gapped site never needs a connection.
The same platform, pointed at the stay.
Nothing here is a hospital product. It is the platform, configured — the same one the rest of this site describes, which is why the deployment that drafts discharge summaries also answers the rest of what your organisation asks of it.
- Systems
- Alongside Epic, Cerner and the rest — EHR & FHIR connectors, 170+ integrations via MCP
- Document intake
- Chart exports, scanned records, watched folders, upload or API
- Reading
- OCR, indexing, medical entity recognition and linking
- Retrieval
- Grounded retrieval across the indexed stay
- Structure
- Knowledge Graph entities and relationships
- Access
- Groups and IAM — least privilege, deny overrides allow
- Oversight
- Human-in-the-loop sign-off and a full audit trail
- Languages
- Correspondence drafted in 120+ languages
- Deployment
- On-premise, private cloud or air-gapped
What hospital teams ask first.
See Exemplary AI in action
Book a demo and we'll show you purpose-built agents grounded in your own knowledge — deployed on your infrastructure.