Unkadlabs/unkad-platform

Qor Af-Soomaali — community platform for building an open, linguist-verified Somali text corpus. Write, translate, validate; peer review + linguist sign-off; releases to Hugging Face. # The Unkad Platform · Qor Af-Soomaali **Open-source infrastructure for building open, quality-controlled text corpora for low-resource languages — by the community that speaks them.** Live instance: **qor.unkad.com** — where Somali speakers are building an open Somali corpus, sentence by sentence. Built and operated by Unkad Labs, a non-profit AI research laboratory, Somalia. *Unkad* is Somali for "creation from nothing." This repo is the whole platform. If your language is missing from the digital record, you can deploy your own instance in under an hour — see Deploy this for your language. ## How it works Contributors work in four modes (named in Somali on our instance): | Mode | | What contributors do | |---|---|---| | **Qor** | write | Respond to prompts — stories, instructions, dialogues — in the target language | | **Turjun** | translate | Translate short source sentences, producing parallel data | | **Guuri** | transcribe | Type up openly licensed / public-domain printed material | | **Hubi** | validate | Review others' submissions: is this correct, natural language? | **Two-tier quality pipeline.** Every submission needs agreement from two independent peer validators (2 approvals → accepted, 2 rejections → rejected, splits escalate). Accepted items then pass to trusted **linguist reviewers** for batch sign-off before they count as verified corpus. Nobody reviews their own work; one vote per person per item, enforced by the database. **Consent first.** Before contributing anything, every contributor completes a three-step onboarding — account, language profile (dialect, region), and an explicit consent step where they accept the open license and choose how to be credited: by name, by pseudonym, or anonymously. Contribution is impossible before consent. **Provenance-complete data.** Every item carries its mode, register (conversational / narrative / instructional / formal / technical), sector (health, education, agriculture, law, media, religion, cul …