AfriLabs
AfriLabs is a network organization that supports innovation hubs, tech communities, and entrepreneurship in Africa. With a presence across 53 countries, AfriLabs plays a key role in fostering the growth of innovative tech solutions that address critical challenges in Africa. By connecting innovators, providing resources, and promoting collaboration, AfriLabs drives entrepreneurship and technological advancements across the continent. The organisation has been a strategic partner in various initiatives aimed at empowering African innovators, including organizing hackathons, incubation programs, and capacity-building initiatives.
Gates Foundation and Meta
The Gates Foundation, through its Gender Equality Digital Connectivity (GEDC) and Digital Public Infrastructure (DPI) teams, is committed to fostering digital inclusion and gender equality across Africa. The GEDC team focuses on developing digital tools and technologies that are sensitive to gender issues, ensuring equitable access to information and services for women.
Meta (formerly Facebook) is a global technology leader, known for its work in advancing artificial intelligence (AI), machine learning, and open-source tools. Meta supports the Llama Impact Grants Program, which seeks to empower innovators and startups to leverage open-source AI models like Llama to address societal challenges, including gender sensitivity and inclusivity, and linguistic diversity.
Together, the Gates Foundation and Meta have partnered to launch a series of programs aimed at improving AI outputs in African languages and ensuring that these solutions are gender-sensitive, linguistically diverse, and culturally relevant.
About the Project
The AI Bias Reduction & Inclusive Data Generation Engine programme is a key initiative designed to engage innovators across Africa in the development of AI-driven solutions that address gender sensitivity, particularly in African languages. The goal is to create innovative AI models that are inclusive and free from gender bias, promoting equitable access to information for both women and men in African communities.
In Phase 1, there was a hackathon where AfriLabs engaged 99 participants, forming 19 country-based teams that developed prototypes over the course of an intensive competition. The top eight winning teams proceeded to apply for the Llama Impact Grants, a fund that supports innovative applications of their open-source AI model, Llama, to address pressing societal challenges.
These teams have now advanced to Phase 2, where four for the eight winning teams have been selected to receive mentor support, to develop a gender sanitization engine that will be analysing and cleaning outputs within 17 pre-determined African languages. Consultancy is a critical component of this initiative, as it provides participants with the guidance needed to refine their work when it comes to the creation of the gender sanitization engine.
Project Objectives
The AI BRIDGE (AI Bias Reduction & Inclusive Data Generation Engine) initiative aims to advance the development of gender-sensitive, inclusive AI tools by engaging African innovators in dataset-level bias mitigation. The primary objectives of Phase 2 of this initiative are to:
- Detect and mitigate gender bias within datasets used for training AI/LLM systems.
- Support dataset curators and AI developers with tools to rewrite biased content while maintaining meaning.
- Promote inclusive AI development across African and global linguistic contexts.
- Enable scalable, modular integration across any AI model pipeline.
This phase builds on earlier outputs from the efforts within the phase one activity; the hackathon and will empower selected teams to build model-agnostic, upstream bias detection and rewriting pipelines that enhance fairness and representation in AI systems within the African context.
Project Methodology
The implementation of AI BRIDGE Phase 2 will follow a structured, modular, and academically rigorous approach. It will blend automated processes with human oversight to ensure gender sensitivity and cultural accuracy in AI training datasets. The methodology is divided into seven critical stages:
a) Data Ingestion
- Source/Curate diverse structured and unstructured datasets across targeted African languages and sectors.
- Align data sources with use cases relevant to gender representation and inclusion.
b) Bias Detection
- Apply hybrid scanning techniques (rule-based logic, machine learning classifiers, and statistical metrics) to identify gender-biased, stereotypical, or exclusionary content.
- Highlight problematic language using linguistic and contextual tagging.
c) Content Analysis
- Conduct contextual tagging of flagged entries, including metadata classification (e.g., severity, stereotype type, grammatical role).
- Organize entries into review-ready clusters for annotation and sanitization.
d) Content Sanitization
- Deploy transformer-based models to suggest rewritten alternatives that are semantically consistent yet bias-free.
- Incorporate a controlled vocabulary for gender-sensitive rewriting.
e) Human-in-the-Loop Validation
- Facilitate expert review from gender specialists, linguists, and cultural advisors to validate rewrites in context-sensitive cases.
- Enable override or acceptance workflows using customized annotation tools.
f) Dataset Finalization
- Finalize sanitized datasets, version-control all changes, and document transformation history for audit and transparency.
- Deliver outputs with full traceability and aligned with ethical AI documentation standards.
g) Feedback and Continuous Learning
- Maintain a feedback loop for rejected or complex cases to improve rule sets and classifier accuracy.
- Integrate feedback into successive model training iterations and guideline updates.
Purpose of the Assignment
AfriLabs is seeking a machine language expert to support the development of AI- driven gender-sensitive tools for African languages. These consultants will play a pivotal role in guiding teams in Phase 2 of the project to ensure that AI outputs are free from gender bias and culturally appropriate. The experts will provide technical guidance, assess linguistic and gender biases in model outputs, and support the integration of inclusive practices within the AI systems being developed. This role requires a strong understanding of AI technologies, gender sensitivity and analysis, linguistic diversity, the African innovation landscape, and familiarity with the nuances of African Sociolinguistic contexts.
Scope of Work
The machine language expert will be responsible for the following:
a) ML Strategy and Model Architecture Advisory:
- Advise participating teams on the overall ML architecture for the AI BRIDGE system, including model selection for bias detection, rewriting, and classification.
- Help teams evaluate trade-offs between rule-based, statistical, and deep learning approaches for different components (e.g., bias identification, language modelling).
- Recommend appropriate supervised, unsupervised, or semi- supervised learning techniques based on data availability and annotation levels.
b) Bias Detection Model Development Guidance:
- Guide the design and fine-tuning of bias detection models, leveraging text classification, sequence labelling, or embedding-based similarity techniques.
- Provide oversight on the use of pre-trained models and their adaptation for fairness-aware NLP tasks.
- Advise on integration of bias scoring and explainability tools (e.g., SHAP, LIME) into detection workflows.
c) Support on Contextual Rewriting and Language Generation:
- Offer guidance on how to use transformer models (e.g., T5, GPT, mBART) for rewriting biased content while maintaining semantic coherence and contextual relevance.
- Help define logic and pipelines for controlled generation, ensuring outputs align with inclusive language norms and domain constraints.
- Recommend strategies for prompt engineering, fine-tuning, or zero- shot classification in low-resource settings.
d) Human-in-the-Loop Learning and Feedback Loops
- Support teams in integrating feedback loops where human reviewer input is used to retrain or calibrate ML models.
- Advise on active learning, few-shot learning, or co-training techniques that enhance performance without needing large labeled datasets.
- Guide the structuring of feedback-informed model updates, versioning, and rollback protocols
e) Performance Evaluation and Monitoring:
- Define evaluation protocols and metrics for ML models, including:
- Bias detection precision/recall
- Rewrite fluency and preservation score
- Model calibration across languages/domains
- Review model evaluation reports and advise on how to interpret results for continuous improvement.
- Suggest appropriate cross-validation, ablation, or benchmarking strategies for multilingual or imbalanced datasets.
f) Tooling and Infrastructure Recommendations:
- Recommend suitable ML frameworks and libraries) for use by teams.
- Guide selection of MLOps tools for experiment tracking, reproducibility, and deployment.
- Ensure that teams use tooling that is aligned with the ethical AI and data protection principles of the programme.
g) Knowledge Transfer and Capacity Strengthening:
- Conduct learning sessions or technical reviews with teams on:
- ML model lifecycle design
- Fairness-aware training techniques
- Model interpretability and documentation
- Mentor participating teams in developing technically sound, context- sensitive, and auditable ML solutions.
Required Qualifications and Experience
- General Qualifications
The expert is expected to possess the following qualifications:
An advanced degree in artificial intelligence, machine learning, computer science, or computational linguistics.
- Solid foundation in the development of natural language processing (NLP) systems, with specific experience working on multilingual or under-resourced language datasets.
b) Specialized Qualifications and Experience
The Consultant should be qualified two or more of the following areas:
- Proven experience designing or evaluating ML models with a focus on fairness, interpretability, and harm reduction.
- Familiarity with frameworks protocols.Deep expertise in NLP model design and application, including transformer- based architectures (e.g., BERT, GPT, T5).
- Experience working with language generation, paraphrasing, or rewriting models for contextual correction tasks.
- Strong experience mentoring or advising interdisciplinary AI teams, especially in early-stage prototyping environments.
- Ability to communicate ML concepts to non-technical stakeholders in ethical, inclusive AI settings.
Proposed/Tentative timeline of deliverables
- Selected Consultants are expected to carry out this assignment for a total of five months (from 7th August 2025 to the 31st of December 2025).
- This will be done remotely
Method Of Application
Applying as an Individual:
- Filled Vendor request form
- Send a Cover Letter, CV or portfolio stating profile and relevant experience.Means of Identification.Financial proposal (This must be in USD)
- Submission of organization reference (i.e Name, email and phone number of organization you have provided similar services)
- A sample of previous review work, preferably related to competitive innovation or grant programs.
Applying as a company:
- Filled Vendor Request Form
- Send your company’s certificate.Company registration document.Cover letter and CV of relevant experience of the applicant or Team Lead.Means of Identification of the team lead.Financial proposal (This must be in USD)
- Submission of organization reference (i.e. Name, email and phone number of organization you have provided similar services) A sample of previous review work, preferably related to competitive innovation or grant programs.
Evaluation Criteria:
- Evaluation of each applicant will be on the basis of the following :
Technical Proposal (65%)
- Submission of the requested documents as outlined in the method of application for each category
- Proof of required qualification and expertise
- Proof of work done in relation to the requested scope
Financial Proposal (35%)
- Financial proposal should be submitted in USD and kindly indicate if the amount proposed is negotiable or not.
- Validity of the financial proposal should be minimum of 6 months.
Note the following:
- Please download and use the attached Vendor information form attached below to submit your application to [email protected] with the email subject: “AI BRIDGE MACHINE LANGUAGE EXPERT”. Applicants should submit their applications on or before COB August 1st, 2025.
- Application should only be submitted if you agree to the renumeration placed within the TOR.
- Kindly find the vendor information form below.
- Please download before use, DO NOT edit the uploaded template.
- Vendor information form Paid Hub members are strongly advised to apply with their company name (If within their thematic areas and expertise).
- Women are strongly encouraged to apply
- Only the shortlisted applicants will proceed to the next stage
