Research

Featured Research Projects

Vociply: Real-Time Voice AI System

Active Research & Development
Deep Learning Indaba 2025

A groundbreaking voice-to-voice agentic system that combines large language models with real-time speech processing to automate business conversations in multiple African languages.

Research Contributions:

  • Novel architecture bridging LLMs with speech processing
  • Creative solutions for multilingual data scarcity
  • Real-time optimization for voice-based AI agents
  • Production-scale deployment of research prototypes

Technical Highlights:

  • End-to-end voice processing pipeline
  • Multilingual natural language understanding
  • Real-time response generation (<200ms latency)
  • Scalable cloud infrastructure

Hypernetwork Optimization Framework

Research Draft
Target: Top-tier AI Conference

Investigating how hypernetworks can optimize ternary neural networks for efficient edge AI deployment, addressing the critical need for AI democratization in resource-constrained environments.

Research Contributions:

  • Can hypernetworks improve ternary network training efficiency?
  • What are the optimal architectures for edge AI deployment?
  • How can we balance accuracy with computational constraints?

Technical Highlights:

  • More efficient AI models for mobile and IoT devices
  • Reduced computational requirements for AI applications
  • Broader accessibility of advanced AI technologies

Multilingual Voice AI Research

Ongoing
Collaboration: Vociply Research Lab

Pioneering research in voice AI systems that work effectively across African languages, addressing fundamental challenges in low-resource language processing.

Research Contributions:

  • Data augmentation for speech data scarcity
  • Transfer learning across related languages
  • Cultural and linguistic adaptation in AI systems
  • Community-driven dataset development

Conference Participation

AMLD Africa 2024

USIU Kenya2024

Publications & Academic Output

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

2025

Maroa, C. et al.

Deep Learning Indaba 2025 [Accepted]

View Paper

Hypernetwork Optimization of Ternary Neural Networks for Edge AI Deployment

In Progress

Maroa, C. et al.

Research Draft [Draft]

Research Pipeline

  • Edge AI democratization tools and methodologies
  • Multilingual conversational AI architectures
  • Audio processing optimization techniques
  • Human-AI interaction paradigm studies

Open Source Contributions

Edge AI Optimization Tools

Available on GitHub | 500+ developers

Open-source utilities for deploying efficient ML models on ARM-based devices, democratizing AI deployment capabilities.

Features:

  • Model compression utilities
  • Performance benchmarking tools
  • Deployment automation scripts
  • Educational resources and tutorials

AI Learning Resources

ARM Ambassador Program | 1000+ learners

Comprehensive learning paths and documentation bridging academic ML research with practical implementation.

Features:

  • Workshop materials and tutorials
  • Code examples and best practices
  • Mentorship programs for emerging researchers
  • Technical documentation and guides