Target clinical speech-assessment workflow

Rubric-grounded assessments.Evidence clinicians can trace.

NeuroTrace connects focused patient audio submission with CTP and PBJ assessment pipelines, using task-specific LLM-as-judge rubrics, deterministic verification, targeted repair, and auditable results for clinician review.

NeuroTrace assessment pipeline: patient audio is transcribed through the approved ASR route, routed to either the Cookie Theft Picture or Peanut Butter and Jelly task-specific rubric judge, deterministically verified and repaired when necessary, then prepared as a traceable result for clinician review.

Patient input

Audio sample

Private transcription

Approved ASR

clearspeechsample

Task-specific route

LLM-as-judge rubrics

CTP

CTP Rubric Judge

Cookie Theft Picture

Task-owned prompt and schema
PBJ

PBJ Rubric Judge

Peanut Butter and Jelly

Task-owned prompt and schema

Evidence check

Deterministic verification

Targeted repair

Clinician review

Traceable result

Patient to clinician

One controlled assessment journey

Purpose-built speech tasks replace the previous generic browser-analysis workflow.

Focused patient submission

Patients record or upload the speech sample requested for their assessment through a clear, distraction-free flow.

Task-specific LLM judges

CTP and PBJ use independent rubrics, prompts, schemas, and explicit assessment graphs instead of one generic analysis.

Verified, traceable results

Deterministic findings, bounded targeted repair, scores, feature vectors, and ordered audit events support clinician review.

Methodology

From patient submission to clinician review

Each task follows an explicit path from intake and transcription through rubric judging, verification, targeted repair, scoring, and an auditable terminal result.

  1. 01

    Patient submission

    Recorded or uploaded audio

  2. 02

    Approved ASR

    Task-scoped private-audio transfer

  3. 03

    CTP / PBJ judges

    Task-owned rubrics and schemas

  4. 04

    Verify and repair

    Deterministic checks and bounded revisions

  5. 05

    Clinician review

    Evidence-linked, auditable outputs

Privacy boundary

Private audio stays local by default

Production audio uses local ASR by default and is not uploaded to a hosted transcription provider. Private transcript transfer to an LLM provider requires explicit task- and provider-specific approval.