Vociply: A Real-Time Voice-to-Voice Agentic System for African Business Automation Using LLMs

DeepLearningIndaba 2025 Workshop AIBF Submission2 Authors

Keywords: Voice-to-voice systems, conversational AI, African languages, code-switching, multilingual NLP, speech-to-speech LLMs, business automation, small medium enterprises, real-time processing, low-latency systems, bandwidth optimization, intent recognition, retrieval-augmented generation, function calling, cultural adaptation, Swahili, Yoruba, emerging markets, digital transformation, ethical AI deployment, data sovereignty, labor displacement, customer service automation, appointment scheduling, African business contexts, network resilience, cross-lingual transfer learning

TL;DR: Vociply is a real-time voice AI system that handles multilingual African business conversations with 87% accuracy and sub-2-second response times, reducing customer service costs by 34% for SMEs across Africa

Abstract: Small and medium enterprises (SMEs) across Africa face significant barriers to adopting advanced customer service technologies due to cost, connectivity, and linguistic diversity constraints. This paper presents Vociply, a real-time voice-to-voice agentic system that leverages large language models (LLMs) to automate business communication for African SMEs. Our system addresses three critical challenges: multilingual support for code-switching scenarios, low-latency operation in bandwidth-constrained environments, and seamless integration with existing business tools. Vociply employs a novel architecture combining speech-to-speech LLMs with retrieval-augmented generation (RAG) and function-calling capabilities to handle customer service, appointment scheduling, and information retrieval tasks. We evaluate our system across English, Swahili, and Yoruba languages, demonstrating 87% task completion accuracy with sub-2s response latency even in low-bandwidth conditions. Our multilingual intent recognition pipeline successfully handles code-switching with 92% accuracy, while maintaining cultural and contextual appropriateness. Through real-world deployment with 15 SMEs across Kenya, Nigeria, and Ghana, we show a 34%reduction in customer service costs and a 28% improvement in customer satisfaction scores. This work contributes a scalable framework for democratizing AI-powered business automation in emerging markets while addressing ethical considerations around labor displacement and data privacy.

Submission Number: 2