Contributor Guide

1. Welcome

Thanks for contributing to DataYetu. We build production-grade African language and voice datasets for AI—starting with East African fintech conversations and wearable/voice commands.

Your submissions are used, with your consent, for commercial AI training and dataset licensing. Read the Terms before you accept.

2. Current pilots

Fintech conversations (text): short Swahili / English / Sheng utterances like sending money, checking balance, paying bills, or disputing a transaction.

Voice (audio): record or upload takes up to 30 minutes each—we can clip shorter segments during labelling. Wearable-style commands, conversation, or other speech are all welcome.

Each project has its own intent list and guidelines. If you hesitate between two intents, pick the closest primary intent; do not invent new labels.

3. How to contribute

Accept Terms (scroll to the end) and acknowledge the Privacy Policy.

Choose Fintech text or Voice command.

Set language and optional city. For voice, pick a device context and record or upload a short clip.

Submit. Your item enters labelling → review → possible QA → approved or returned.

Only approved samples enter exportable datasets. Raw and reviewed data are not mixed in releases.

4. Quality expectations

Be natural and locally authentic. Prefer real phrasing over stiff textbook Swahili when the project calls for Sheng/code-switching.

Do not paste private chats that contain other people’s phone numbers, ID numbers, or full transaction secrets.

For voice: speak clearly when possible; mark noise honestly during annotation. Unusable audio may be rejected without pay.

One primary intent per sample. Label only entities that are clearly present.

5. Pay and withdrawals

Pay varies by project, modality, duration/quality, and role. Check the Payments tab for the current rate card and balances.

Contributor pay is typically credited when a sample is approved. Annotators/reviewers may earn on tracked active time.

Withdrawals target USDC on Base once you meet the threshold and set a wallet. Manual review may apply in early pilots.

Rejected or forfeited items do not pay. Gaming the system can lead to account suspension.

6. If you also annotate or review

Label quickly but carefully: intent → entities → sentiment (and voice fields when shown).

Reviewers approve, correct, or reject. Corrections should fix the label, not rewrite the contributor’s meaning without cause.

Follow weekly guideline updates when published.

7. Help

Questions: hello@datayetu.com

Legal: /legal/terms and /legal/privacy

This guide is operational help, not a contract. The Terms control if there is a conflict.