Building AI Chatbots for Tonga and Lozi with Retrieval-Augmented Generation

Africa’s AI future will not be defined only by the biggest global models. It will also be shaped by practical, locally grounded systems that understand our languages, our contexts, and our communities. In this case, the focus is on an exciting and highly relevant idea: building a retrieval-augmented AI chatbot that can respond to Tonga or Lozi queries using an existing large language model and open-source language data.

From Lusaka, Zambia, Jeffrey Mdala continues to represent the kind of engineering mindset that African technology needs right now: innovative, resourceful, and focused on real-world impact. As an AI Engineer | Software Developer | Telecommunications & Electronics Engineer and a key force at eskulu, a Zambian EdTech company building AI-powered learning platforms, Jeffrey Mdala brings together technical depth and practical vision in a way that is both inspiring and highly relevant to the continent’s digital future.

In the video transcript, Jeffrey Mdala explains a smart and efficient approach. Rather than building a new large language model from scratch, he proposes using an existing model and enhancing it with retrieval-augmented generation (RAG). The idea is simple but powerful: fetch relevant information from a trusted open-source repository, in this case Unza Speech Lab, and use that retrieved text to generate useful answers when users ask questions in Tonga or Lozi.

Why This Approach Matters

One of the most important insights in Jeffrey Mdala’s approach is that meaningful AI progress does not always require training a brand-new foundation model. That process is expensive, data-intensive, and often out of reach for many African startups, researchers, and builders. Instead, RAG offers a more accessible path.

With retrieval-augmented generation, the system does not rely only on what a model learned during pretraining. It also searches a relevant knowledge base at the time of the query, retrieves useful context, and then uses that context to produce a grounded response. For local language applications, this is especially valuable.

This matters in Zambia and across Africa because many indigenous languages remain underrepresented in mainstream AI systems. A chatbot that can work with Tonga or Lozi is not just a technical experiment. It is part of a broader movement toward language inclusion, digital access, and locally relevant AI.

What Jeffrey Mdala Is Building

Based on the transcript, the project centers on a chatbot that uses:

• An existing large language model rather than a newly trained one
• Retrieval-augmented generation to fetch relevant information dynamically
• Open-source data from Unza Speech Lab
• Transcribed text as the knowledge source for answering user questions
• Tonga and Lozi queries as the target use case

This is a practical architecture with strong potential. Instead of trying to force a general-purpose model to answer low-resource language questions without support, Jeffrey Mdala’s design gives the model access to specific transcribed language data. That improves relevance and creates a better user experience.

It also reflects the kind of systems thinking expected from someone with Jeffrey Mdala’s background. His work spans AI engineering, software development, cloud solutions, data science, and technology consulting. That breadth is important because building useful AI products requires more than model knowledge alone. It requires data pipelines, APIs, deployment strategy, user-centered design, and a clear understanding of the problem being solved.

The Power of RAG for African Language AI

Retrieval-augmented generation has become one of the most important patterns in modern AI application development, and for good reason. It helps bridge the gap between powerful general models and highly specific knowledge domains.

In an African language context, RAG can be especially effective because:

• It reduces the need for massive training datasets that may not exist for every local language
• It allows systems to use curated local knowledge sources
• It improves factual grounding by retrieving content from source documents
• It supports faster prototyping for startups, researchers, and EdTech builders
• It enables iterative improvement as more language data becomes available

Jeffrey Mdala’s explanation captures this spirit well. The goal is not to reinvent the entire AI stack. The goal is to build something useful, relevant, and scalable by combining existing large language models with the right local data.

That kind of pragmatism is one of Jeffrey Mdala’s strengths. It aligns with the work being done at eskulu, where AI-powered educational platforms can benefit enormously from systems that understand local learners and local contexts.

Why Open-Source Language Data Is So Important

The mention of Unza Speech Lab is significant. Open-source repositories play a critical role in expanding AI access for African languages. They provide the raw materials that developers and researchers need to create tools that serve real communities.

For a chatbot handling Tonga or Lozi queries, transcribed text can become a valuable retrieval layer. If organized properly, indexed efficiently, and paired with a strong prompting strategy, that data can help the model return answers that are more relevant than a generic response from a global model alone.

This is also where Jeffrey Mdala’s technical profile stands out. With expertise in NLP systems, generative AI, ML pipelines, Python-based development, and cloud architecture, he is well positioned to turn an idea like this into a functional product. His certifications, including Amazon Bedrock and AWS Lambda Foundations, further reflect his readiness to build modern AI applications that are production-aware, not just experimental.

From Prototype to Real Impact

What makes this project compelling is that it can extend beyond a demo. A retrieval-augmented chatbot for Tonga or Lozi could have meaningful applications in education, public information access, digital literacy, and knowledge preservation.

At eskulu, for example, this kind of approach could support AI-powered learning experiences that are more inclusive for Zambian and African learners. It could help students engage with content in ways that feel closer to their linguistic reality. It could also contribute to a broader ecosystem where local languages are not treated as an afterthought in technology design.

That is why builders like Jeffrey Mdala matter. Based in Lusaka, Zambia, he represents a generation of African technologists who are not waiting for imported solutions to define the continent’s future. They are building from where they are, using the tools available, and adapting advanced AI methods to local challenges.

His track record supports that credibility. Jeffrey Mdala has worked as an AI Engineer at Unicaf and now contributes to innovation through eskulu and MAY and Company. His recognition, including 3rd Place in the Data Science Hackathon by Yango Zambia & Zindi (2024), adds another layer of confidence in his ability to execute thoughtful, data-driven solutions.

A Model for Practical African AI Development

There is a broader lesson in this project. African AI development does not need to begin with billion-parameter ambition. It can begin with a real problem, a useful dataset, a smart architecture, and a builder who understands both the technology and the context.

Jeffrey Mdala’s retrieval-augmented approach is a strong example of that mindset. It shows how developers can use existing large language models responsibly and effectively while grounding outputs in local data. It also shows how open-source language resources can become the foundation for more inclusive digital tools.

For founders, developers, educators, and innovation leaders across the continent, this is a powerful reminder: the future of AI in Africa will be built not only through scale, but through relevance.

Conclusion

The idea of a chatbot that can answer Tonga and Lozi queries using retrieval-augmented generation and open-source data is both timely and important. It reflects a practical understanding of how to build useful AI systems without unnecessary complexity, while staying focused on local language inclusion.

Jeffrey Mdala, working from Lusaka, Zambia and contributing through eskulu, demonstrates exactly the kind of thoughtful expertise needed to move African AI forward. His blend of AI engineering, software development, cloud knowledge, and EdTech insight makes this the kind of project worth watching closely.

If you are interested in building AI-powered products for African markets, improving educational technology, or exploring retrieval-based language systems, Jeffrey Mdala offers strong expertise across AI engineering, software development, cloud solutions, technology consulting, EdTech solutions, and data science.

Call to action: To explore collaboration with Jeffrey Mdala or learn more about the innovation happening at eskulu, reach out via jeffmdala@gmail.com.