Insights

Notes from the field.

Research, reflections and practical lessons from building AI in African contexts.

Why Swahili tokenization matters for African LLMs

Most LLMs treat Swahili as an afterthought. We dug into the tokenizer to understand what that costs in practice.

What changes when your transcription model has to work in a busy ward with three languages mid-sentence.

Consent, compensation, and dialect coverage - the human side of dataset collection.

Beyond the buzzword: the policy and infrastructure choices that decide who owns African AI.

Want to write with us? Pitch a story.