Divvun & Giellatekno — Sámi Language Technology Infrastructure
Norway
Since 2004, Norway's Divvun and Giellatekno research groups at UiT have built keyboards, spellcheckers and machine translation for all eight …
South Africa · Johannesburg · See the South Africa profile · See the Johannesburg profile
Top 67% 56/100 · Ask Evidence Copilot about this practice
A Johannesburg startup built Vulavula, a speech-to-text and text-analysis tool for isiZulu, Sesotho and Afrikaans, so telecom and bank call centres can serve African-language speakers without forcing them into English.
Founded in Johannesburg in 2022 by computer scientist Jade Abbott and biomedical engineer Pelonomi Moiloa — who had earlier co-founded the grassroots NLP research collective Masakhane — Lelapa AI builds compact, resource-efficient language models purpose-built for under-resourced African languages, rather than adapting large Western-language models. In November 2023 it launched Vulavula, a commercial API offering speech-to-text transcription and named-entity/text analysis for isiZulu, Sesotho, Afrikaans and English, aimed at telecom and financial-services contact centres that need to serve customers in their own language.
The company raised $2.5 million in seed funding in February 2024 from investors including Mozilla Ventures and Google chief scientist Jeff Dean, and in 2024 published InkubaLM, an open small language model trained for five African languages, documenting its low-resource training methodology. As an early-stage venture with a single funding round and roughly three years of operating history, its long-term financial sustainability is not yet proven, and no independent third-party evaluation of Vulavula's transcription accuracy or downstream customer outcomes has been published to date — the evidence so far is product and launch coverage rather than a rigorous impact study, which this assessment reflects honestly.
Read the full analysis: https://lelapa.ai/products/vulavula/
Implementation detail (cost, timeline, staffing, conditions for success) is not yet available for this practice.
Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.
Where this practice's information was retrieved from, and when.
Norway
Since 2004, Norway's Divvun and Giellatekno research groups at UiT have built keyboards, spellcheckers and machine translation for all eight …
Rwanda
Kigali-based Digital Umuganda crowdsourced one of the largest open Kinyarwanda voice datasets (1,700+ hours from 840+ contributors) via Mozilla Common …
Togo
Togo built a fully digital, mobile-money cash-transfer system using satellite imagery and phone data to target its poorest citizens during …
India
A free voice forum lets Gujarati farmers who can't read record questions and hear expert answers by phone. A randomized …
Open full copilot Grounded in cited practices — always check the sources.