Main site

Healthcare AI agent · Clinical documentation

Clinical-Documentation & Ambient-Scribe AI Agent

Capture the consented encounter, draft the note into your own templates and hand the clinician a draft to edit — with unsupported lines marked, subgroup evaluation, audit trails and nothing filed before signature.

Typical delivery
4–6 weeks
Deployment
Your stack
Note signature
Clinician only
After launch
Agent Care

What this agent does

Drafts the note, does not sign it

In

  1. 01

    Record the encounter only after consent is captured on the client's terms, on the agreed device and channel.

  2. 02

    Read the chart the clinician already has open — problem list, medications, allergies, recent results and the last note.

Reason

  1. 03

    Separate the speakers and keep every statement attached to whoever made it — clinician, patient, carer or interpreter.

  2. 04

    Draft the note into the client's own template and smart-phrase library, section by section, rather than a house format.

  3. 05

    Tie each drafted line back to the point in the encounter it came from, so text the encounter does not support is visible.

Decide

  1. 06

    Detect content in the draft that nobody said and hold it back instead of letting it read like the rest of the note.

  2. 07

    Mark thin audio, crosstalk, interpreted encounters and weakly supported sections for the clinician to read closely.

Out

  1. 08

    Present the draft with suggested orders, problem-list changes and coding hints — unplaced, unposted and uncommitted.

  2. 09

    Retain the consent, the transcript, the draft, every edit made before signature and the signature event itself.

Product statement

The agent drafts. The clinician edits, attests and signs — nothing reaches the chart unsigned, no order is placed and no billing code is committed by the agent.

Example workflow

One encounter, end to end

  1. 1Consent capturedRecorded on the client's terms before any audio is taken, with a refusal path that does not delay the visit
  2. 2Encounter capturedAudio from the agreed device, alongside the chart the clinician already has open
  3. 3Speakers separatedEvery statement kept attached to the person who made it — clinician, patient, carer or interpreter
  4. 4Draft written to your templateThe client's own sections and smart phrases, with each line tied back to the point it came from

No human action required

Stages 1 to 4 run without a person in the loop — capture, speaker separation and drafting finish before the clinician is asked to read anything. Nothing is filed anywhere in that stretch.

5DecisionSplits on how much of the draft the encounter supports
Fully supported draft

Reaches the clinician as a clean draft.

Unsupported or thin

Reaches the clinician marked line by line.

Clinician review

The draft is held beside the encounter it came from, with unsupported lines marked and the suggested orders and codes still unplaced.

Edit · Attest and sign · Discard the draft

Signed — handed back

  1. 6Filed on signatureWritten to the chart only when the clinician signs; suggested orders and codes stay unplaced until a clinician acts
  2. 7Outcome evaluatedUnsupported content caught, omissions, editing before signature, coding agreement and subgroup results

Edits

Every edit made before signature is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Signing or attesting a note.
    • Placing an order or a prescription.
    • Committing a diagnosis or a billing code.
    • Writing up an examination that was not performed.

  • Automation boundary

    Agent acts unaided

    • ✓Capture the encounter once consent is recorded on the client's terms.
    • ✓Separate speakers and keep every statement attached to the person who made it.
    • ✓Draft the note into the client's own template and smart-phrase library.
    • ✓Mark unsupported lines and hold suggested orders and codes unplaced for the clinician.

    Write actions run only inside the approval boundaries agreed during implementation. Signature, orders and codes are not among them.

    • Adding history the encounter did not contain.
    • Amending or addending a note already signed.
    • Recording without consent, or overriding a refusal.
  • Changing templates, retention or coding rules.

Example output

One drafted note, annotated

Everything the agent drafts is attached to the point in the encounter it came from.

Draft output · single clinic encounter

Illustrative example
Encounter
Follow-up visit, in person
Capture
Consent recorded first
Speakers
3 separated
Draft handed to
Clinician, unsigned
Line support
97%
Orders and codes
Suggested, not placed

As received

The encounter as it was captured, and the consent recorded before capture began — nothing on this side is drafted, summarised or inferred.

Evidence used

  • Consent on file
  • Speaker-separated transcript
  • Client note template

Why nothing is filed yet

The agent has no signature authority. Until a clinician attests, the draft sits outside the chart and the suggested orders sit unplaced.

What the score decides

Line support marks which lines to read closely, not whether a note is safe to sign. A clinician signs either way.

Action

EditAttest and signDiscard the draft

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every consented encounter

In the room, on the agreed device

03Capture & drafting

Apply the client's own template

Use the approved note templates, smart phrases, specialty sections and the chart already on file.

01Clean draft

Take the typing out of the visit

The clinician leaves the room with a drafted note in their own format instead of an evening of catch-up documentation.

02Clinician review

Put the doubtful lines in front of the clinician

Unsupported and thin sections are marked in the draft rather than buried in text that reads as fluently as the rest.

04

Build an evidence trail

Retain the consent, the speaker-separated transcript, the draft, every edit made before signature and the signature event — on both paths.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • EHR & note

    • Epic
    • Oracle Health
    • MEDITECH
    • athenahealth
    • note write-back

  • Orders & results

    • Order sets
    • e-prescribing
    • Result and referral queues
    • care plans

  • Capture & speech

    • In-room mic
    • mobile app
    • telehealth
    • Speech pipeline
    • interpreter channel

Agent

Clinical documentation & ambient scribe

  • Captures on consent
  • Drafts the note
  • Holds for signature

  • Templates & coding

    • Note templates
    • smart phrases
    • Coding and CDI
    • problem-list rules

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the chart

Each control wraps the one inside it. A draft clears every layer before a clinician reads it, and the signature sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn documentation to the existing workflow if evaluations or production signals degrade.Roll back
L5TraceabilityRecord the consent, transcript, draft, every edit and the signature event.Record
L4Signature gateFiling, orders and codes wait for a clinician to edit, attest and sign.Gate
L3Consent and captureNothing is recorded or kept beyond the consent given for that encounter.Verify
L2Template guardrailsRestrict the draft to the client's approved sections, phrases and defaults.Restrict
L1Support checkingLines the encounter does not support are held back and marked.Hold back
Model coreDraft note proposed — sectioned text, suggested orders, coding hints and line support
L1 – L2
Decide whether a drafted line may stand
L3
Decides what may be recorded at all
L4 – L5
Keep signature with a clinician and the trail intact
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the draft against the encounter — not only how it reads.

Coverage runs the whole depth of the workflow, and every layer is cut by slice. The suite is built first to catch content the encounter never contained.

Surface — the draft the clinician opens

  1. E1Final-output evaluationIs the note faithful to the encounter it was drafted from?
  2. E2Step-level evaluationWere the speakers separated and the right template sections filled?
  3. E3Tool evaluationDid it read and write the correct patient, encounter and note?
  4. E4Fabrication and omissionHow much unspoken content survives, and how much said content is lost?
  5. E5Slice evaluationHow does faithfulness change across specific encounter cohorts?
  6. E6Business outcomeHow much was edited before signature, and did the coding hints agree with the coder?

Floor — the signed note in the chart

Failure modes

Where each failure originates in the agent

Seven failure modes plotted against the five stages of the agent lifecycle.

Agent lifecycle

Direction of processing →

  1. 01 · Capture

    1 mode

    • CD-01

      Accented speech degrades

      Word error rises with accent, crosstalk or room noise.

    Stage records

    The consented audio, the room and the open chart

  2. 02 · Transcription

    2 modes

    • CD-02

      Negation flipped

      A denial in the room becomes a finding in the note.

    • CD-03

      Dose or laterality misheard

      A strength, frequency or side is transcribed wrong.

    Stage hears

    Words, speakers, negation, doses and laterality

  3. 03 · Drafting

    2 modes

    • CD-04

      Content nobody said

      Unspoken detail appears in the draft, including exam findings.

    • CD-05

      Said but not drafted

      Something the clinician stated is missing from the note.

    Stage writes

    The note sections, orders and coding hints proposed

  4. 04 · Review / signature

    1 mode

    • CD-06

      Signed without reading

      A fluent draft is attested with its errors intact.

    Stage returns

    The draft the clinician edits, attests and signs

  5. 05 · Change / Version

    1 mode

    • CD-07

      Silent template regression

      A template or model change moves note content unnoticed.

    Stage tracks

    Model, prompt, template and coding-rule changes

  • Sev-1 · wrong content reaches the signed chart
  • Sev-2 · the draft misleads the clinician
  • Sev-3 · the draft degrades, more is rewritten
See the complete Healthcare failure-mode catalogue

Affected slices

Faithfulness is not spread evenly across encounters

Aggregate note quality can look acceptable while a small number of encounter cohorts carry most of the unsupported lines, most of the rewriting before signature and most of the risk that something nobody said gets signed. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Interpreted and non-English visits5.8%3.6×Review
Accented and non-native speech4.2%2.6×Review
Noisy rooms and telehealth audio3.1%1.9×Watch
Routine single-problem follow-ups1.0%0.6×Normal

Bar: lift vs. routine follow-up baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A line nobody said becomes a test before it becomes a fix

One piece of unsupported text is not closed when the clinician deletes it. It becomes a held-out encounter the next drafting change has to clear.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Unsupported lines, omissions or editing before signature move in a cohort.

  2. 02

    Diagnose

    Traced to the audio, the speaker separation, the transcript or the drafting step.

  3. 03

    Improve

    The prompt, template or support rule is changed, re-approved and version-linked.

  4. 04

    Verify

    Re-run against held-out encounters from the affected cohort, including the ones that failed.

  5. 05

    Learn

    That encounter becomes a regression case and the changed rule enters the documentation runbook.

  6. Learn → Detect

    The return edge. A template or coding-rule change is a documentation-integrity change — it is signed off before it ships, not after.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, capture, speech and drafting, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02EHR note, order and coding API assessment.Week 1, Week 2
  3. 03Consent capture, recording policy and audio retention.Week 1, Week 2
  4. 04Note-template and smart-phrase mapping.Week 2
  5. 05Capture on the agreed devices and channels.Week 2, Week 3
  6. 06Speech pipeline and speaker separation.Week 3
  7. 07Draft generation against client templates.Week 3, Week 4
  8. 08Order and coding suggestion handling.Week 4
  9. 09Fabrication and omission evaluation suite.Week 4, Week 5
  10. 10Accent, language and interpreted-visit testing.Week 5
  11. 11EHR note write-back and signature workflow.Week 5, Week 6
  12. 12Observability, deployment and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne specialty, one siteProductionProduction EHR integrationAdvancedMulti-specialty / multi-site
Introduced at Pilot
Drafted notes from consented capture✓✓✓
Consent capture and retention policy✓✓✓
Clinician attestation and signature✓✓✓
Baseline faithfulness evaluation✓✓✓
Introduced at Production
Client templates and smart phrases—✓✓
Speaker separation in multi-speaker rooms—✓✓
Order and coding suggestions, unplaced—✓✓
EHR note write-back—✓✓
Observability and subgroup evaluation—✓✓
Introduced at Advanced
Coding and CDI integration——✓
Multi-specialty and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on specialties and templates, EHR, order and coding integrations, encounter volume, consent and retention requirements, approval controls and deployment.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your consent, recording and audio-retention policy

    Consent capture, recording policy and retention rules

    Week 1

  2. 02Your note templates, smart phrases and specialty sections

    Note-template and smart-phrase mapping

    Week 2

  3. 03Access to EHR note, order and coding APIs

    Epic and Oracle Health note, order and coding write-back

    Week 2

  4. 04Recordings or transcripts you are cleared to use

    Speech pipeline, speaker separation and the drafting baseline

    Week 3

  5. 05Notes that went wrong, including the ones that got signed

    Evaluation suite, regression cases and failure-mode testing

    Week 4

  6. 06The languages, accents and interpreter arrangements in your rooms

    Accent, language and interpreted-visit testing

    Week 5

  7. 07Named clinicians who will edit and sign the drafts

    Signature workflow, then supervised drafting and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Phases are drawn over the weeks they actually occupy. Week 5 carries both the subgroup evaluation and the first drafts a clinician signs.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Consent, recording and audio-retention policy signed off

W2Note templates, API assessment and the first capture channel

W3Speech pipeline, speaker separation and drafting into your sections

W4Fabrication and omission evaluation suite and regression encounters

W5Accent and interpreted-visit slices, write-back, first signed drafts

W6Clinicians sign live drafts in clinic, then Agent Care handover

Reading the band
Each bar covers the weeks its work is named in. The accent and interpreted-visit slices sit in week 5, cut from the first signed drafts.
At the end of W6
Drafts have been edited and signed by the clinicians who own the notes, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Healthcare AI agent

Build a documentation agent around the note your clinicians sign.

Show us your note templates, your consent and recording policy, and how a note reaches the chart today. We'll draft one encounter into your own template and agree what may never reach the chart unsigned.

Typical delivery
4–6 weeks
Deployment
Your EHR and documentation stack
Note signature
Clinician, always
After launch
Agent Care available

Nestack Agents · Clinical documentation & ambient scribe

AGT-HC-02 · Agent Care available after launch

Build a clinical-documentation agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your systems

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.