AfriHealth AI: Unified 3-in-1 Voice Care Suite

MODULE 1

Frontline Community Triage Agent

Supporting community health workers (CHWs) with voice-driven Amharic-English triage documentation.

Noise Suppression: Active Mode: Amharic-English Code-Switch

Select Clinical Audio Preset or Record Live

Ready to record or run pre-loaded sample

Recording language

Audio processed via Intron Sahara v2.5 ASR Engine

Intron Sahara v2.5 Code-Switched ASR Stream Confidence: --%

Click a preset above or tap the microphone to begin code-switched speech recognition...

Extracted Medical Entities

Awaiting audio processing...

Triage Classification

Pending Processing

Actionable referral recommendation will be generated automatically.

EVIDENCE REVIEW

ASR Benchmark and Clinical Validation

The matrix below is a fixture/demo view. It is not a real model ranking; use the measured validation summary and linked reports for performance claims.

Fixture benchmark limitation

The displayed matrix uses embedded hypotheses and illustrative case values. Its 0% WER-style result is reproducible as a software fixture, but it is not independent audio inference or proof of production superiority. The "All Cases" row is calculated live as the average of the six per-case rows below, not a separately hardcoded figure.

Measured clinical validation

15 reviewed Amharic-English recordings: 56.38% mean WER, 44.33% target-term recall, and 6 critical-term-miss cases. Clinician review is mandatory; autonomous care is not supported.

Active Case Filter:

Duration: 71.8s Code-Switching: 63.5% Amharic / 36.5% English

Raw Ground-Truth JSON payload (GOLD-ETH-001)

WER % vs CER % Comparison (Lower is Better)

Aggregated Corpus

Clinician Human-In-The-Loop (HITL) Scores

Speech AI Model	WER	CER	Factuality	Safety

Clinician Review Notes:

Intron Sahara v2.5 maintains accurate phonetic grounding for local disease descriptions without hallucinatory translations.

Live Intron Sahara v2.5 Benchmark

Upload or record one reviewed Amharic-English sample and measure the configured Intron provider. OpenAI and Gemini can be added later.

Ready

Sample audio

No recording

Verified reference transcript (optional)

Reference mode

Models: Intron OpenAI Gemini

Live Levenshtein Edit-Distance Calculator

Formula: WER = (S + D + I) / N

Reference Verbatim Ground Truth ($T_{gold}$)

Hypothesis ASR Model Output ($T_{hyp}$)

Calculated WER: --% Calculated CER: --%

AfriHealth AI Unified Care Suite

Frontline Community Triage Agent

Select Clinical Audio Preset or Record Live

Extracted Medical Entities

Triage Classification

Voice Intake & EHR Integration Engine

Doctor-Patient Code-Switched Conversation

Auto-Filled EHR Medical Chart

Structured SOAP Encounter & Diagnostic Output

Post-Care Voice Follow-up

Outpatient Call Simulator

Live Post-Care Protocol Analyzer

Real Voice Follow-up Call

Outbound Voice Interaction Transcript

Parsed Status

Escalation Queue

ASR Benchmark and Clinical Validation

Fixture benchmark limitation

Measured clinical validation

WER % vs CER % Comparison (Lower is Better)

Clinician Human-In-The-Loop (HITL) Scores

Live Intron Sahara v2.5 Benchmark

Live Levenshtein Edit-Distance Calculator

System Setup & Hardware Diagnostics

Frontline Community Triage Agent

Select Clinical Audio Preset or Record Live

Extracted Medical Entities

Triage Classification

Voice Intake & EHR Integration Engine

Doctor-Patient Code-Switched Conversation

Auto-Filled EHR Medical Chart

Structured SOAP Encounter & Diagnostic Output

Post-Care Voice Follow-up

Outpatient Call Simulator

Live Post-Care Protocol Analyzer

Real Voice Follow-up Call

Outbound Voice Interaction Transcript

Parsed Status

Escalation Queue

ASR Benchmark and Clinical Validation

Fixture benchmark limitation

Measured clinical validation

WER % vs CER % Comparison (Lower is Better)

Clinician Human-In-The-Loop (HITL) Scores

Live Intron Sahara v2.5 Benchmark

Live Levenshtein Edit-Distance Calculator

AfriHealth AI Product Strategy

Purpose and product direction

1. Product modules

2. Evidence and governance

3. Privacy and safety

4. Project team