mabera/nigeria-multilingual-health-model

Nigeria Multilingual Health Model

AutoScientist Challenge 2026 | Language Category

Author: Hussein Adeiza (mabera)
Role: Licensed Environmental Health Officer, Abuja Nigeria
Base Model: Mixtral 8x7B
Fine-tuned with: AutoScientist by Adaption Labs

Model Description

This is a LoRA adapter fine-tuned on a multilingual Nigerian public health Q&A dataset covering WASH, malaria and poverty topics in 5 languages — English, Hausa, Yoruba, Igbo and Nigerian Pidgin.

Training Data

• Source: Original Q&A pairs grounded in DHS Nigeria, OPHI MPI and UNDP HDR
• Dataset: 29 Q&A pairs expanded to 11,200 rows via Adaptive Data
• Languages: English, Hausa, Yoruba, Igbo, Nigerian Pidgin
• Quality improvement: 33.3% (Grade C → B)
• Kaggle: https://www.kaggle.com/datasets/yunusahusseinadeiza/nigeria-multilingual-health-qa-dataset

Training Metrics

• Win rate: 60% adapted vs 40% base model
• Base model: mistralai/Mixtral-8x7B-Instruct-v0.1
• Method: LoRA — House Special + Hallucination mitigation
• Dataset quality: 6.0 → 8.0 (+33.3% improvement)

Languages Covered

Language	Speakers	Domain
English	Global	WASH, Malaria, Poverty
Hausa	70M+	WASH, Malaria, Poverty
Yoruba	45M+	WASH, Malaria, Poverty
Nigerian Pidgin	75M+	WASH, Malaria, Poverty
Igbo	30M+	WASH, Malaria

Why This Matters

Nigerian languages — Hausa, Yoruba, Igbo and Pidgin — are spoken by 200+ million people yet are massively underrepresented in AI health datasets. A farmer in Kano, a mother in Ibadan, a trader in Onitsha — none of them can access public health AI in their language. This model changes that.

Credits

Powered by Adaptive Data — Adaption Labs
AutoScientist Challenge 2026 | Language Category