Pelonomi Moiloa

Pelonomi Moiloa is building language AI from a premise the industry has too often ignored: people should not surrender ownership of their languages to gain access to modern technology. Her work pairs efficient models with a practical economy of community rights.

For Pelonomi Moiloa, language is not merely training material; it is memory, commerce and the right to participate in public life. As co-founder and chief executive of Lelapa AI, she has built that conviction into systems designed for African languages and constrained computing environments. In 2025 she co-authored the peer-reviewed Esethu Framework, which couples community-led dataset curation with a commercial licence that waives fees for African entities and requires proceeds from non-African commercial use to be reinvested in further language data and fairly paid local work. Its proof of concept, the Vuk’uzenzele isiXhosa Speech Dataset, provides a research-accessible corpus for speech recognition while keeping stewardship close to its creators. The company also opened InkubaLM to an international compression challenge: more than 490 data scientists from 61 countries participated, and the strongest submissions reduced model size by as much as 75% while preserving performance. An affiliated 2025 convening report recorded the company’s claim that its isiZulu transcription and translation system used roughly 70% less data than comparable models and still outperformed available alternatives. Moiloa’s achievement is architectural and civic at once: she is showing that resource efficiency and community agency can be sources of technical advantage, not concessions at the edge of global AI.