MoraaOntita/Zuhura-Voice-Agent

Zuhura Voice Agent is a collection of production-ready AI voice agent prompts for debt collection in Kenya. It enables natural Kenyan Swahili conversations, handles transcription errors, supports M-Pesa payments, adapts across collection stages, and includes smart escalation, time conversion, and context-aware dialogue management. # Zuhura Voice Agent A collection of AI-powered voice agent prompts and system instructions designed for debt collection operations in the Kenyan market. These prompts provide comprehensive guidelines for voice-based AI agents to handle natural Swahili conversation with debtors while managing complex interactions with professionalism and efficiency. ## Project Overview This project contains system prompts and operational guidelines for voice AI debt collection agents operating in Kenya. The agents are designed to handle conversations in Swahili (Kenyan dialect), adapt to regional language patterns, and manage debt collection calls with three distinct modes: initial calls, follow-ups, and late follow-ups. ## Project Structure ``` ├── agents/ │ ├── initial_prompt.md │ ├── version_1.md │ └── version_2.md ├── Truncated/ │ ├── follow_up_TRUNCATED.md │ ├── initial_TRUNCATED.md │ └── late_followUp_TRUNCATED.md └── README.md ``` ## Core Content ### **Debt Collection Agent Prompts** (`agents/`) - **Version**: 1.2 Multi-Mode - **Purpose**: AI voice agent system prompts for conducting debt collection calls - **Available Versions**: - **initial_prompt.md** - Original comprehensive multi-mode prompt (initial, follow-up, late follow-up) - **version_1.md** - Modified version - **version_2.md** - Further optimized version - **Call Modes**: - **Initial calls**: Inform debtors of overdue balance and secure payment commitment - **Follow-up calls**: Follow-up on previous calls with familiar yet professional tone - **Late follow-up calls**: Address broken promises and escalate when necessary - **Key Features**: - Multi-mode call type handling - Audio transcription error recovery - Context-based inference for unclear speech - Swahili time system conversion (6-hour offset) - Currency and country validation - M-Pesa payment integration - Smart escalation protocols ## Key Language & Technical Features ### Swahili Language Support - **Primary Language**: Swahili (Kenya …