Skip to main content
NGPT mini sample · updated 2026-06-23

Inspect the data shape before you commission a Nigerian-language set.

A tiny synthetic public sample covering A/B preference ranking, translation, and voice-read metadata. It is intentionally not a training corpus; it proves schema, consent, QA, and provenance fields so buyers can review the pipeline before procurement.

5
Sample rows
4
Approved examples
1
Rejected transparency row
0
Raw PII fields
Synthetic
Illustrative rows only

Consent spine

Rows expose ai_data_reuse consent scope and provenance hashes. Production exports require live consent and exclude revoked rows.

RLHF-ready

Preference rows include prompt, answer A/B, choice, and rationale; tie/both-bad choices are retained for QA but can be dropped from DPO exports.

QA visibility

The sample intentionally includes one rejected row so buyers can see exclusion reasons, not just polished approved records.

Sample Rows

License: 9jatesters-consented-data-licence-v1. Commercial use by contract.

IDKindLanguagePrompt / answerQAConsentRegion
ngpt-pref-001ab_preferencepidgin

User asks: My transfer never show for receiver account. Wetin I fit do?

Choice A: A is safer and gives practical next steps. B may cause double debit and is bad financial advice.

approvedai_data_reuseLagos
ngpt-pref-002ab_preferencehausa

User asks for a polite reminder to repay a small loan.

Choice A: A is polite and culturally safer. B is threatening and should be rejected for tone.

approvedai_data_reuseKano
ngpt-trans-001translationyoruba

Translate: Please keep your receipt until the payment is confirmed.

Response: Jowo pa risiti re mo titi di igba ti won ba fi jerisi owo sisan naa.

approvedai_data_reuseOyo
ngpt-voice-001voice_readpidgin

Read aloud: I transfer the money for OPay, but e never enter.

approvedai_data_reuseRivers
ngpt-reject-001voice_readenglish

Read aloud: My network is slow this evening.

rejected

Silent recording; excluded from exports and payout.

ai_data_reuseAbuja