Yetu
Clinical documentation system under development

From clinical signal to accountable record.

Yetu is engineering a consent-visible, offline-resilient capture-to-record system for Indian clinical workflows. The architecture keeps encounter identity, source lineage and clinician authorship connected from the first control event to the approved record.

Explore the systemWork with Yetu
Capture state
Deliberate and visible
Connectivity model
Offline resilient
Final authority
Responsible clinician
Reference architecturecapture / record
  1. 01
    Wearable controlDeliberate capture / visible state
    local
  2. 02
    Phone edgeConsent / encounter / encrypted queue
    bound
  3. 03
    Clinical intelligenceLanguage / facts / chronology
    versioned
  4. 04
    Provenance layerField / source / model / template
    traceable
  5. 05
    Clinician approvalReview / correct / sign / release
    human
Documentation assistance onlyNo autonomous charting

The systems problem

A note is only as dependable as the chain that produced it.

Speech-to-text solves one component. Clinical documentation also has to preserve who initiated capture, which encounter it belongs to, what survived an interruption, where each field came from and who accepted responsibility for the final record.

Capture stateEncounter identityProcessing lineageClinical approval
System sequence / 05 stages

Follow one encounter through the system.

Select a stage to inspect the handoff, the mechanism and the variable that needs to hold.

Stage 01

Deliberate capture

Input
A clinician starts, pauses and stops the session deliberately.
System action
The wearable records visible control states and divides the encounter into ordered, time-stamped chunks.
Handoff
A bounded capture session, ready to be attached to one encounter.
What to test
Control-state fidelity, missed capture and acoustic completeness.

Architecture

The hard part sits between capture and approval.

Yetu treats documentation as a distributed clinical system. Each layer has a specific job, an explicit failure mode and a variable that can be measured.

  1. 01

    Capture

    Deliberate, visible session control

    A physical state machine starts, pauses and closes an authorised recording. Time-stamped chunks enter encrypted local storage before transfer.

    Unintended, missed or incomplete capture

    Activation fidelity, acoustic completeness, false and missed capture

  2. 02

    Bind

    Consent and encounter identity

    The phone-edge layer is intended to bind device, clinician, purpose and encounter before processing begins.

    A recording reaches the wrong encounter

    Binding accuracy, recovery behaviour, consent comprehension

  3. 03

    Reconcile

    Offline integrity across interruption

    Ordered chunks, integrity hashes and retry-safe synchronisation preserve the session when connectivity drops or the application is interrupted.

    Loss, duplication or silent corruption

    Recovery rate, duplicate prevention, manifest integrity, latency

  4. 04

    Structure

    Source-linked clinical drafting

    Selected-language transcription and clinical structuring are designed to keep each drafted field linked to its source span, model and template version.

    Fluent text hides an omission or unsupported statement

    Clinical-term accuracy, critical omissions, unsupported content, correction effort

  5. 05

    Approve

    Clinician authorship and controlled release

    The clinician reviews uncertain spans, corrects the draft and signs the record before any configured export or institutional write-back.

    Automation becomes unaccountable charting

    Review completion, edit history, attestation and export audit

Read the complete platform architecture
Industrial design study of a white clinician pendant with a central control and lanyard
Industrial design studyIntended form factor, not a final production device

Why a wearable

The hardware must earn its place in the workflow.

A wearable is useful only if deliberate physical control, a consistent capture position and local buffering improve real clinical work. Yetu intends to compare the complete system with a phone-only baseline rather than assume the form factor wins.

  • Visible intentA physical action and patient-visible state can make recording easier to recognise.
  • Hands-busy captureBody-worn control may reduce phone handling during procedures and rehabilitation work.
  • Local continuityAn encrypted buffer can preserve the authorised session when connectivity or application state fails.

Comparator variables include missed capture, correction time, acoustic completeness, comfort, charging burden, hygiene and social acceptability.

Engineering validation

The weakest indispensable layer sets the maturity of the system.

Hardware, synchronisation, language models, review workflow and governance have to work together. Product-level claims should follow integrated tests, not component demonstrations.

01

Device and connectivity

Can every authorised session survive interruption intact?

Acoustics, power, local buffering, transfer resumption, corruption and duplicate prevention
02

Language and grounding

What changes across speciality, accent and code switching?

Word error rate, clinical-term accuracy, speaker attribution, omissions and unsupported statements
03

Clinical workflow

Does the complete system reduce work without hiding risk?

Time to approved note, correction burden, interruption count and required-field completeness
04

Human factors

Do clinician and patient understand the recording state?

Activation recognition, consent comprehension, comfort, hygiene and accidental operation
05

Governance

Can access, retention and release be reconstructed later?

Role controls, audit completeness, deletion behaviour, export policy and controlled failure

Candidate defensibility

Value can accumulate in the implementation.

The microphone, language model and template are not the whole product. Defensibility can emerge from how the system controls state, reconciles failure and turns governed corrections into reusable validation assets.

01

Consent-bound state orchestration

Device identity, clinician, encounter, purpose and recording state move through one explicit control model.

02

Offline reconciliation

Integrity checks, idempotent retry and duplicate prevention preserve the record chain through unreliable connectivity.

03

Field-level provenance

A drafted field can retain its source segment, processing history, version, clinician edit and approval event.

04

Indian clinical-language evaluation assets

Ethically governed, selected-language and code-switching tests can create a compounding technical asset over time.

Build with Yetu

A platform worth shaping before the architecture hardens.

Early partners can influence the workflow, comparator, safety thresholds and integration model while every important decision can still be tested.

01

Clinicians

Define the narrow workflow, the fair comparator and which errors could change care.

02

Hospitals and teaching centres

Shape consent, governance and integration before a pilot reaches routine operations.

03

Research partners

Build adjudication protocols and ethically governed Indian clinical-language evaluation assets.

04

Technical and strategic partners

Resolve coupled problems across hardware, resilient edge computing, language, security and deployment.

Interested in a clinical, research or engineering partnership?Tell us the workflow or technical problem you want to test.
Start a conversation