BantuNomics is not a dataset you buy once. It is a living platform that starts at the absolute fundamentals and continuously, masterfully curates everything built on top of them — deepening and widening over time.
The syllable inventory is the operating alphabet of a Bantu language — like the 26 letters, but the real atomic units native speakers actually read by. Every layer above (words, verbs, tone, meaning, health, conversation) decomposes back into syllables, and nothing above can be validated without the alphabet beneath it. So the foundation is laid first — and it is already the broadest layer: 459 released operating alphabets across the 500+ Bantu languages.
The result is a broad, solid foundation with layers climbing on top of it at different rates — the build frontier always moving upward. What you see below is where the platform is today; it grows continuously.
The closed set of a language's legal syllables — its actual alphabet. Everything above decomposes back into it, so it is built first and it is built broadest.
Every verb lemma expands into its full paradigm through the 9-slot verb equation (39 verb types) — each cell a real, syllable-validated surface form.
Same spelling, different meaning, different tone. Meaning-anchored code-switch takes capture the layer flat text drops and a model can't self-certify.
Body parts and clinical phrases — consented, care-critical, and captured in both a clean take and the real Bantu→English code-switch a patient actually speaks.
Two-mode numeral audio — clean Bantu and the code-switch — where ASR and language ID most often break.
The build never finishes. Next layers rising on the foundation:
See what you can use and license today, or — as an evaluation partner and above — where it's going next.
The catalogue → The roadmap