Livejollof-mnk-en · v0.1
Mandinka ↔ English translation, first release
Fine-tuned from NLLB-200 on community-contributed parallel sentences. The first machine-translation model targeting Mandinka publicly available for direct use.
Open the live demoJollofLab is an open research effort building parallel datasets and translation models for languages the major systems don't serve well — beginning with Mandinka, Wolof, Fula, Joola, Soninke, and Manjago.
What's shipping right now and what's next on the roadmap.
Fine-tuned from NLLB-200 on community-contributed parallel sentences. The first machine-translation model targeting Mandinka publicly available for direct use.
Open the live demoThe current generation of large language and translation models works well for a few dozen languages. For thousands of others — spoken by hundreds of millions of people — the data needed to train them simply doesn't exist in machine-readable form.
JollofLab builds the missing pieces: parallel sentence datasets, peer-review workflows for quality control, and open-weight translation models. Everything we ship is released under permissive licenses so other researchers can build on it.
Our work focuses first on the languages of Senegambia, where the founding team is rooted. The platform is designed to expand to any low-resource language community willing to contribute.
Bantaba AI is the first effort to assemble a complete artificial-intelligence stack for the languages of the Senegambia region — Mandinka, Wolof, Fula, Joola, Soninke, and Manjago. Together these languages are the daily means of communication for more than twenty million people across The Gambia, Senegal, and Guinea-Bissau.
Despite that reach, they remain almost entirely absent from modern AI systems. As artificial intelligence becomes the primary interface for services, information, and economic opportunity, that absence is no longer just a technical gap — it is a structural barrier to inclusion.
The initiative will deliver an integrated suite of technologies — machine translation, speech recognition, and text-to-speech — supported by large-scale curated datasets and deployed through open platforms and developer APIs.
The foundation is already operational. Our live Mandinka–English translation model demonstrates both technical feasibility and organizational capability. Bantaba AI is a strategic scale-up into regional infrastructure, not a proposal in the abstract.
Cycles through phrases the Mandinka model has been trained on. Switch target languages to see contributor-provided translations side by side.
Hello, how are you?
Released as we reach data and quality thresholds. Each model card lists the training base, the language pair, and the data scale at release time.
First release. Trained on community-contributed parallel sentences.
Data collection in progress. Awaiting reviewer threshold.
Open for contributors and reviewers.
These are the languages with active or planned dataset work. The roster grows as new contributor communities come on board.
Language not listed? Propose one →
Foundational work informing our approach to low-resource translation. Our own technical reports will be added here as they're published.
Researcher or institution interested in collaborating? contact@jolloflab.com →
Contributors translate sentences in languages they speak fluently. Reviewers validate the work of others. Both roles are open and self-paced — the platform tracks your reputation as your contributions are accepted.