Datasets and benchmarks
Domain and language corpora, plus evaluation benchmarks for African languages, covering the gap IrokoBench measured across 17 of them.
Capabilities
One engine, five stages, every kind of African data.
Nyansora runs a single pipeline from memory, field and shelf to model-ready data. One engine, applied across language, culture, climate, agriculture, health and finance.
The pipeline
From an elder’s memory to a training set.
01
Collect
Paid, consent-based collection through community networks, covering speech, text and field records in African languages. Every item carries provenance from the day it is gathered.
02
Digitize
OCR and transcription lines for knowledge that exists only on paper or tape, across public records, print and broadcast archives.
03
Label
Trained annotation teams turn raw collections into training-grade datasets, transcribed, translated, tagged, reviewed and quality-scored to the standard global labs already pay for.
04
Preserve
A governed corpus of cultural and indigenous knowledge, held so the communities who produced it are named, credited and paid when it is used.
05
Serve
Licensed datasets, APIs and evaluation benchmarks for model builders, plus insight products mapping crop, price, flood, health and demand trends.
Who we serve
Five groups, one underlying asset.
Insight products
Crop and price trends, flood-risk maps and demand patterns for agribusiness, banking, insurance, health and education.
The cultural corpus
Licensed to institutions, media houses, publishers and tourism boards for curricula, film, festivals and games.
Commissioned programmes
Digitization and structured-data programmes for ministries and national archives.
Institutional access
Universities and R&D institutions subscribe for access to the corpus under research terms.
Standards
What ships with every dataset.
Every record carries provenance: who provided it, under what consent, and for which permitted uses. Every release is reviewed before it ships.
Cultural and indigenous collections are governed with their source communities, with attribution and a revenue share written into the licence. Operations comply with Ghana’s Data Protection Act, 2012, and with the African Union Data Policy Framework.
Priority
Preservation is the work that cannot wait.
A language disappears roughly every two weeks, and each one takes its oral archive with it. Manuscripts burn. Elders die. Every other dataset can be collected again next year. This one cannot.
Data is the new gold of the AI age. Africa should be the miner and the owner, not the mine.