Offerings
ቋንቋAmharic for “language”
Data for every stage of training.
From the corpus a model learns a language from, to the experts who teach it to reason in that language. Native from the start, never translated in.
Pre- & mid-training data
Native datasets in dialect-specific language. Large, clean corpora written by native speakers and labelled by dialect, so your model learns how the language is actually used: at home, at work and in print.
- i.Books & long-form writingOut-of-copyright Arabic books, digitised and cleaned. Deep, structured text across centuries and genres.
- ii.Dialect corporaNatively written text in regional varieties, tagged by dialect so you can balance your mix.
- iii.Company & domain dataLicensed, sector-specific sources in the language of real businesses and professions.
- iv.Custom sourcingNeed a dialect, domain or format we don’t list? We’ll source it to your spec.
In the catalog now
60,000
Arabic books, with the authority to license them and full OCR at 99% accuracy
Printed books whose text was largely missing from the data today’s models learn from, now available as clean, machine-readable text. Reach out to request a sample, cut to what you are training or evaluating.
Post-training data
Human domain experts who work in your target language. They build the evals that tell you whether your model is good enough, and the training data that makes it better, with a focus on under-resourced languages.
We have a strong network of Arabic-speaking professionals across dialects and across domains: medicine, finance, law, management and more.
- i.Custom evaluationsExpert-written, dialect-aware test sets and rubrics for your use case.
- ii.Post-training datasetsDemonstrations, preference rankings and graded reviews for fine-tuning.
- iii.Native domain expertsArabic-speaking doctors, financial professionals, lawyers, managers and other specialists, across dialects, sourced and vetted through our own pipeline.
How we work
- i.
Scope
Tell us the language, dialect, domain and training stage.
- ii.
Sample
We deliver a sample so you can check quality before committing.
- iii.
Deliver
Data and expert work at scale, tracked in the Logha app.
- iv.
Iterate
We refine with you as your model and evals improve.
Questions from AI teams
Is your data translated?
No. Everything is written natively by speakers of the language and dialect. That’s the point.
How is data delivered?
Through your workspace in the Logha app, where you can track progress and download deliveries.