Odunayo Ogundepo

Blog

Thoughts on NLP, multilingual systems, evaluation, and building AI systems that work in the real world.

Building Multilingual NLP Systems: Lessons from African Languages

Working with low-resource languages presents unique challenges in NLP. In this post, I share key insights from developing multilingual systems for African languages, including data collection strategies, model selection, and evaluation frameworks that can transfer to other low-resource settings.

Understanding RAG Systems: Beyond Basic Retrieval

Retrieval-Augmented Generation has become a cornerstone of modern LLM applications. This post explores advanced RAG patterns, including hybrid retrieval strategies, relevance assessment, and how to build robust systems that know when they don't know.

Thoughts on Model Evaluation in Production

Academic benchmarks don't always translate to production success. Drawing from experience building personalized AI systems, I discuss practical evaluation strategies, monitoring approaches, and how to close the loop between offline metrics and real-world performance.

Open Source and Research: Building KarantaOCR

Reflections on building an open-source OCR system for African languages. I share the technical journey, from data collection and model architecture to community collaboration and deployment strategies that make language technology more accessible.

Built from scratch by me and Claude :)