EqualyzAI crafts Africa-centred AI

African-centric AI startup EqualyzAI has unveiled a suite of AI tools that it says are designed to democratise AI innovations for the continent.

EqualyzAI has developed AI models built with African linguistic patterns and cultural nuances at their core, addressing a gap where major AI platforms struggle with non-Western languages and contexts.

The company is addressing a major equity gap in technology by working to ensure AI can understand, process, and make decisions in African languages and contexts.

EqualyzAI’s agentic AI product for education, uLearn, was recently showcased as an example of the use of Meta’s large language model (Llama2) to build high-impact solutions. EqualyzAI says its technology addresses the critical exclusion of over 1.5-billion people from meaningful AI engagement.

“Our mission goes beyond technology – it’s about fundamental digital inclusion,” said Dr. Olubayo Adekanmbi, EqualyzAI co-founder and a Gates Foundation Global Grand Challenge winner.

“When today’s AI systems can’t understand your language or context, you’re essentially locked out of the AI revolution. EqualyzAI ensures African languages, cultures, and contexts are central to global AI evolution.”

EqualyzAI’s approach to hyperlocal agentic AI is based on building foundational datasets that enable its localised models to learn effectively. The company uses a Pan-African data collection network to implement a linguistics-guided, people-focused, and data-owner-informed methodology.

The start-up says many African-centric AI models today rely on scraped online data, religious texts, radio broadcasts, and generic datasets, which fail to capture undocumented conversational styles, dialectal variations, and accent patterns.

EqualyzAI has developed an inclusive data collection, validation, and enrichment ecosystem for a sustainable data pipeline through a crowd-validated approach, where it engages and incentivises native speakers across communities to generate and validate audio by dialect; read and record words in local dialects; review and improve translation accuracy; translate complex code-mixed sentences; label images to enhance multimodal AI capabilities.

EqualyzAI says it enables:

• Businesses to engage with African consumers in their native languages.
• Educational institutions to create AI-powered learning materials in local dialects.
• Healthcare providers to improve medical communication across language barriers.
• Government agencies to analyze citizen feedback in multiple languages.
• Financial services to deliver inclusive banking solutions to underserved populations.

Dr. Adebara, co-founder of EqualyzAI, says: “What sets our technology apart is that our small language models consistently outperform much larger global models when it comes to African languages. For example, our AfroSLM 1.0 beats Google Translate and other popular LLMs in contextual translation of finance-related conversations in Nigerian languages.”

EqualyzAI says its key innovations include:

1. Equalyz Crowd: A platform designed to collect hyperlocal datasets needed to build AI solutions that can be trained to understand and speak many African languages, ensuring inclusivity and representation.
2. AfroSLM 1.0: An SLM optimised to improve financial literacy in Yoruba and Hausa, empowering users with accessible financial education in their native languages. Our SLMs outperformed leading large language models in understanding African linguistic nuances and contexts. For instance, our AfroSLM 1.0 beats Google Translate and other popular LLMs in the contextual translation of finance-related conversations in Nigerian languages.
3. uLearn: An interactive learning tool helping teachers generate lesson plans, notes, and quizzes in both English and local languages, transforming education across Africa.
4. AgentZero: A street-smart customer service chatbot understands emerging slang and informal expressions in delivering relevant and practical customer service.
5. LLMiner: A tool that helps organisations and governments extract actionable insights from years of archived audio, video, or scanned documents. By structuring these datasets into query-able tabular data, LLMiner enables seamless, conversational-styled data interaction – like chatting with someone on WhatsApp.

EqualyzAI says its approach represents a significant shift from current AI development methodologies, which have largely overlooked the linguistic diversity of Africa’s 2,000+ languages. By building models from the ground up with African languages in mind, EqualyzAI is creating a more inclusive and equitable AI landscape.

EqualyzAI has published a white paper titled How Small Language Models (SLMs) Are Revolutionizing Localized AI Innovations for Social Good in Africa. The findings explore the transformative potential of SLMs in addressing Africa’s distinct challenges and advancing AI accessibility.