sociable systems.
HubGrieVoice OverviewLive Agent DemoModular ArchitectureComprehensive DeckWhatsApp Integration SpecUSSD Integration Spec

Modular Architecture

System architecture diagram — intake channels, processing layers, outputs.

GrieVoice

Multi-Channel Grievance Collection Architecture

📱 Intake Channels

🎙️

Voice Agent

Real-time empathic voice conversation with emotion detection

Live
📝

Web Form

Text-based submission with GPS location capture

Built
💬

WhatsApp Voice

Async voice messages transcribed & processed

Planned
📞

USSD

Menu-driven, works on any phone, no data required

Planned
→
AUDIO / TEXT
→

⚙️ Processing Layer

AI Processing Stack

Hume EVI / Whisper
Speech Recognition
Claude Sonnet
Field Extraction & Classification
GPT-3.5 / Gemini
Translation (PT/SW → EN)
Keyword Analysis
Urgency Detection

Structured Output

Submitter name/contact
Incident date & location
People involved
Category (customizable)
Description
Urgency level
Source channel
→
STRUCTURED DATA
→

📊 Backend & Outputs

🗄️

Supabase

PostgreSQL database with unified grievance table + audio storage

Live
📈

Dashboard

Filter by date, category, urgency, source. Export to CSV/Excel

Live
🔔

Notifications

Email/SMS alerts for high-urgency cases, daily digests

Planned
📑

Reporting

IFC-compliant reports, trend analysis, resolution tracking

Planned

Modular Design Philosophy

All intake channels feed into the same processing layer and unified database. New channels can be added without modifying existing infrastructure. Categories, languages, and urgency rules are configurable per deployment. The system is designed for contexts where workers have varying levels of literacy, technology access, and comfort with different communication methods.

Per-conversation cost: $0.26 (enterprise) to $0.65 (pilot) | Phone integration: +$0.17/call