Details
Mission and objectives
UNDP works in 170 countries and territories to eradicate poverty while protecting the planet. We help countries develop strong policies, skills, partnerships and institutions so they can sustain their progress
The primary and overarching objective of United Nations Development Programme in Kenya is the eradication of poverty in the context of sustainable development, including the pursuit of the Sustainable Development Goals, and promotion of United Nations fundamental principles. A core dimension to the work of UNDP in Kenya is on Democratic Governance given the national focus on governance reforms. UNDP supports the country’s efforts towards achieving the Vision 2030 Political Pillar, which envisions a democratic system that is issue-based, people-centered, results oriented and accountable to the public. This Political Pillar gears to transform the country’s political governance across five strategic areas; The Rule of Law, Electoral and Political Processes, Democracy and Public Service Delivery, Transparency and Accountability, Security Peace Building and Conflict Management. These strategic areas are anchored in the Constitution, promulgated in August 2010 marking a major milestone in the democratic journey of Kenya and set a new threshold in terms of people-centred development.
The primary and overarching objective of United Nations Development Programme in Kenya is the eradication of poverty in the context of sustainable development, including the pursuit of the Sustainable Development Goals, and promotion of United Nations fundamental principles. A core dimension to the work of UNDP in Kenya is on Democratic Governance given the national focus on governance reforms. UNDP supports the country’s efforts towards achieving the Vision 2030 Political Pillar, which envisions a democratic system that is issue-based, people-centered, results oriented and accountable to the public. This Political Pillar gears to transform the country’s political governance across five strategic areas; The Rule of Law, Electoral and Political Processes, Democracy and Public Service Delivery, Transparency and Accountability, Security Peace Building and Conflict Management. These strategic areas are anchored in the Constitution, promulgated in August 2010 marking a major milestone in the democratic journey of Kenya and set a new threshold in terms of people-centred development.
Context
In Kenya, UNDP is developing an integrated Artificial Intelligence Programme in partnership with the Ministry of Information, Communications and the Digital Economy (MICDE) and the Technopolis Development Authority (TDA). The programme connects national compute infrastructure, public-sector capability and use cases, responsible AI and data governance, and the innovation ecosystem, creating a practical pathway through which development, institutional and enterprise challenges can be identified, assessed and translated into appropriate AI-enabled solutions that can be responsibly developed, adopted and scaled.
A key component is the operationalisation of national AI compute infrastructure hosted at the Technopolis Development Authority, together with the governance, access and allocation mechanisms, technical capabilities and sustainability arrangements required for its effective use. The infrastructure supports priority use cases emerging from Government, startups and the innovation ecosystem, universities and research institutions, and UNDP Kenya programmes. Following a national open call and a two-stage business and technical assessment, a cohort of priority use cases has been selected for supported solution development.
To operationalise this model, UNDP Kenya is mobilising complementary specialist expertise across the AI solution lifecycle. The team works within the overall programme architecture under the guidance of the Artificial Intelligence Programme Manager, who retains overall portfolio leadership, prioritisation and accountability.
This UN Volunteer assignment contributes to the implementation phase of the Artificial Intelligence Programme by leading the data, modelling and evaluation layer of the priority AI-enabled solutions supported through the programme, and by ensuring that each solution is built on data that can support it and is demonstrably better than a credible baseline.
A key component is the operationalisation of national AI compute infrastructure hosted at the Technopolis Development Authority, together with the governance, access and allocation mechanisms, technical capabilities and sustainability arrangements required for its effective use. The infrastructure supports priority use cases emerging from Government, startups and the innovation ecosystem, universities and research institutions, and UNDP Kenya programmes. Following a national open call and a two-stage business and technical assessment, a cohort of priority use cases has been selected for supported solution development.
To operationalise this model, UNDP Kenya is mobilising complementary specialist expertise across the AI solution lifecycle. The team works within the overall programme architecture under the guidance of the Artificial Intelligence Programme Manager, who retains overall portfolio leadership, prioritisation and accountability.
This UN Volunteer assignment contributes to the implementation phase of the Artificial Intelligence Programme by leading the data, modelling and evaluation layer of the priority AI-enabled solutions supported through the programme, and by ensuring that each solution is built on data that can support it and is demonstrably better than a credible baseline.
Task description
Under the direct supervision of the UNDP Artificial Intelligence Programme Manager, and working alongside the Senior AI Solution Architect & MLOps Lead, the UN Volunteer will undertake the following tasks:
Assess data readiness and define the modelling approach for the selected AI solutions.
• Review the pre-engineering dossiers for programme-selected use cases and assess whether the available data can realistically support the intended AI capability, documenting data gaps, usable proxies, mitigation options and any recommendation to re-scope or defer a use case.
• Conduct data quality, coverage, representativeness, labelling, provenance and licensing or consent assessments for each selected use case, in coordination with data owners and the Responsible AI Specialist.
•Define the training, validation and test data strategy for each solution, including sampling, data splits, labelling protocols and annotation guidance, class imbalance handling, and augmentation or synthetic-data approaches where appropriate.
• Select modelling approaches that are proportionate to the problem and the data, choosing between conventional machine learning, NLP, computer vision, generative or large language model, retrieval-augmented or hybrid approaches on the basis of data, accuracy, cost, latency and explainability requirements rather than technology novelty.
• Define, with use-case owners, a credible non-AI or simple baseline for each solution and the model-quality thresholds that must be met for the solution to be considered technically viable.
•Provide the modelling requirements needed by the Senior AI Solution Architect & MLOps Lead, including model interfaces, input and output contracts, and dependency and resource profiles, so that architecture and compute design reflect actual model needs.
Develop, adapt and fine-tune models for the five (5) selected use cases.
• Build and maintain data preparation, cleaning, feature engineering, prompt engineering and retrieval pipelines required to produce reliable training and evaluation datasets.
• Lead the development, adaptation, fine-tuning, parameter-efficient tuning, distillation, quantisation or other optimisation of models for five (5) selected priority use cases, including an initial proof-of-capability workload where directed by UNDP and where data and access readiness permit.
• Run structured experimentation, including hyperparameter search, ablation studies and prompt or retrieval configuration testing, with tracked experiments, versioned datasets and reproducible training runs.
• Optimise the accuracy, size, latency and inference cost of models against the provisioned on-premise NVIDIA GPU-based compute environment, in coordination with the Senior AI Solution Architect & MLOps Lead, and document the trade-offs of material modelling choices.
• Deliver trained model artefacts, inference code, configuration and interface specifications to the engineering team in the formats, standards and repositories agreed for integration into end-to-end solutions.
• Technically guide the junior AI developers and other programme-mobilised contributors on data preparation, labelling, training and evaluation work packages allocated to them, and review their modelling outputs.
Evaluate, benchmark and validate model performance.
• Develop and apply an evaluation protocol for each selected solution, including baseline comparison, task-appropriate quality metrics, held-out and where feasible real-world test sets, and user or operational acceptance measures defined with use-case owners.
• Conduct failure-mode and error analysis for each model, including performance on edge cases, under-represented groups and out-of-distribution inputs, and translate findings into prioritised model-improvement actions.
• Test explicitly for overfitting, data leakage, spurious correlation and evaluation-set contamination, and state clearly whether each model performs materially better than the agreed simpler baseline, recommending redesign or discontinuation where it does not.
• For generative and large language model based solutions, evaluate groundedness, factual accuracy, hallucination rate, prompt robustness and output consistency using documented and repeatable methods.
• Work with the Responsible AI Specialist to carry out model-level bias, fairness, robustness and privacy testing, and produce the model-level assurance evidence required by the programme, in alignment with the national Data Protection Act and UNDP's data, AI and digital principles.
• Define model-drift, degradation and retraining indicators and thresholds and provide them to the Senior AI Solution Architect & MLOps Lead for implementation in monitoring.
Build internal modelling capability and complete handover.
• Mentor junior AI developers and designated internal technical staff through paired modelling sessions, code and notebook review, experiment reviews, debugging support and technical clinics, with explicit transfer of modelling and evaluation knowledge rather than reliance on individual implementation.
• Produce model documentation (model cards) and dataset documentation (data sheets) for each supported solution, covering intended use, training data, evaluation results, known limitations and monitoring indicators.
• Prepare retraining and model-update playbooks so that internal teams can refresh datasets, retrain, re-evaluate and extend the models after the assignment ends.
• Prepare a consolidated cross-solution modelling report covering approaches used, benchmark results, reusable components, data lessons and recommendations for subsequent AI solution development.
• Provide a final handover of training and evaluation code, data preparation pipelines, model artefacts and configuration in formats approved by UNDP, subject to applicable data, intellectual property and licensing restrictions.
Furthermore, UN Volunteers are required to:
• Strengthen their knowledge and understanding of the concept of volunteerism by reading relevant UNV and external publications and take active part in UNV activities (for instance in events that mark International Volunteer Day);
• Be acquainted with and build on traditional and/or local forms of volunteerism in the host country;
• Reflect on the type and quality of voluntary action that they are undertaking, including participation in ongoing reflection activities;
Results/Expected Outputs
•Data readiness assessment, modelling approach and evaluation protocols - Data readiness and feasibility assessment completed for each programme-selected use case, covering data quality, coverage, representativeness, labelling, provenance and licensing or consent, with identified gaps and mitigation options; documented modelling approach and
justification per use case; agreed baseline definition and model-quality thresholds established with use-case owners; evaluation protocol specifying metrics, test sets, failure-mode analysis approach and acceptance measures; and model interface, dependency and resource requirements submitted to the Senior AI Solution Architect & MLOps Lead.
• Training datasets, model baselines and reproducible experimentation — Prepared and versioned training, validation and test datasets, labelling protocols and data preparation, feature or prompt pipelines delivered for the selected solutions; initial model baselines or proof-of-capability results documented against the agreed baseline; experiment tracking and reproducible training runs established, including dataset and model versioning; modelling work packages allocated to junior AI developers with technical briefs, review cadence and an issues and remediation log; and first-round error and failure-mode analysis completed with prioritised improvement actions.
• Five developed and evaluated models — For five (5) selected use cases: models developed, adapted or fine-tuned and evaluated against the agreed criteria, with inference code, configuration and interface specifications provided to the engineering team for integration; solution-specific model evaluation reports including baseline comparison, metric results and error and failure-mode analysis; explicit technical viability determination per solution with a recommendation to proceed, redesign or discontinue; model-level bias, fairness, robustness and privacy testing evidence produced jointly with the Responsible AI Specialist; and documented model optimisation for the available GPU and compute environment.
Assess data readiness and define the modelling approach for the selected AI solutions.
• Review the pre-engineering dossiers for programme-selected use cases and assess whether the available data can realistically support the intended AI capability, documenting data gaps, usable proxies, mitigation options and any recommendation to re-scope or defer a use case.
• Conduct data quality, coverage, representativeness, labelling, provenance and licensing or consent assessments for each selected use case, in coordination with data owners and the Responsible AI Specialist.
•Define the training, validation and test data strategy for each solution, including sampling, data splits, labelling protocols and annotation guidance, class imbalance handling, and augmentation or synthetic-data approaches where appropriate.
• Select modelling approaches that are proportionate to the problem and the data, choosing between conventional machine learning, NLP, computer vision, generative or large language model, retrieval-augmented or hybrid approaches on the basis of data, accuracy, cost, latency and explainability requirements rather than technology novelty.
• Define, with use-case owners, a credible non-AI or simple baseline for each solution and the model-quality thresholds that must be met for the solution to be considered technically viable.
•Provide the modelling requirements needed by the Senior AI Solution Architect & MLOps Lead, including model interfaces, input and output contracts, and dependency and resource profiles, so that architecture and compute design reflect actual model needs.
Develop, adapt and fine-tune models for the five (5) selected use cases.
• Build and maintain data preparation, cleaning, feature engineering, prompt engineering and retrieval pipelines required to produce reliable training and evaluation datasets.
• Lead the development, adaptation, fine-tuning, parameter-efficient tuning, distillation, quantisation or other optimisation of models for five (5) selected priority use cases, including an initial proof-of-capability workload where directed by UNDP and where data and access readiness permit.
• Run structured experimentation, including hyperparameter search, ablation studies and prompt or retrieval configuration testing, with tracked experiments, versioned datasets and reproducible training runs.
• Optimise the accuracy, size, latency and inference cost of models against the provisioned on-premise NVIDIA GPU-based compute environment, in coordination with the Senior AI Solution Architect & MLOps Lead, and document the trade-offs of material modelling choices.
• Deliver trained model artefacts, inference code, configuration and interface specifications to the engineering team in the formats, standards and repositories agreed for integration into end-to-end solutions.
• Technically guide the junior AI developers and other programme-mobilised contributors on data preparation, labelling, training and evaluation work packages allocated to them, and review their modelling outputs.
Evaluate, benchmark and validate model performance.
• Develop and apply an evaluation protocol for each selected solution, including baseline comparison, task-appropriate quality metrics, held-out and where feasible real-world test sets, and user or operational acceptance measures defined with use-case owners.
• Conduct failure-mode and error analysis for each model, including performance on edge cases, under-represented groups and out-of-distribution inputs, and translate findings into prioritised model-improvement actions.
• Test explicitly for overfitting, data leakage, spurious correlation and evaluation-set contamination, and state clearly whether each model performs materially better than the agreed simpler baseline, recommending redesign or discontinuation where it does not.
• For generative and large language model based solutions, evaluate groundedness, factual accuracy, hallucination rate, prompt robustness and output consistency using documented and repeatable methods.
• Work with the Responsible AI Specialist to carry out model-level bias, fairness, robustness and privacy testing, and produce the model-level assurance evidence required by the programme, in alignment with the national Data Protection Act and UNDP's data, AI and digital principles.
• Define model-drift, degradation and retraining indicators and thresholds and provide them to the Senior AI Solution Architect & MLOps Lead for implementation in monitoring.
Build internal modelling capability and complete handover.
• Mentor junior AI developers and designated internal technical staff through paired modelling sessions, code and notebook review, experiment reviews, debugging support and technical clinics, with explicit transfer of modelling and evaluation knowledge rather than reliance on individual implementation.
• Produce model documentation (model cards) and dataset documentation (data sheets) for each supported solution, covering intended use, training data, evaluation results, known limitations and monitoring indicators.
• Prepare retraining and model-update playbooks so that internal teams can refresh datasets, retrain, re-evaluate and extend the models after the assignment ends.
• Prepare a consolidated cross-solution modelling report covering approaches used, benchmark results, reusable components, data lessons and recommendations for subsequent AI solution development.
• Provide a final handover of training and evaluation code, data preparation pipelines, model artefacts and configuration in formats approved by UNDP, subject to applicable data, intellectual property and licensing restrictions.
Furthermore, UN Volunteers are required to:
• Strengthen their knowledge and understanding of the concept of volunteerism by reading relevant UNV and external publications and take active part in UNV activities (for instance in events that mark International Volunteer Day);
• Be acquainted with and build on traditional and/or local forms of volunteerism in the host country;
• Reflect on the type and quality of voluntary action that they are undertaking, including participation in ongoing reflection activities;
Results/Expected Outputs
•Data readiness assessment, modelling approach and evaluation protocols - Data readiness and feasibility assessment completed for each programme-selected use case, covering data quality, coverage, representativeness, labelling, provenance and licensing or consent, with identified gaps and mitigation options; documented modelling approach and
justification per use case; agreed baseline definition and model-quality thresholds established with use-case owners; evaluation protocol specifying metrics, test sets, failure-mode analysis approach and acceptance measures; and model interface, dependency and resource requirements submitted to the Senior AI Solution Architect & MLOps Lead.
• Training datasets, model baselines and reproducible experimentation — Prepared and versioned training, validation and test datasets, labelling protocols and data preparation, feature or prompt pipelines delivered for the selected solutions; initial model baselines or proof-of-capability results documented against the agreed baseline; experiment tracking and reproducible training runs established, including dataset and model versioning; modelling work packages allocated to junior AI developers with technical briefs, review cadence and an issues and remediation log; and first-round error and failure-mode analysis completed with prioritised improvement actions.
• Five developed and evaluated models — For five (5) selected use cases: models developed, adapted or fine-tuned and evaluated against the agreed criteria, with inference code, configuration and interface specifications provided to the engineering team for integration; solution-specific model evaluation reports including baseline comparison, metric results and error and failure-mode analysis; explicit technical viability determination per solution with a recommendation to proceed, redesign or discontinue; model-level bias, fairness, robustness and privacy testing evidence produced jointly with the Responsible AI Specialist; and documented model optimisation for the available GPU and compute environment.
At Impactpool we do our best to provide you the most accurate info, but closing dates may be wrong on our site. Please check on the recruiting organization's page for the exact info. Candidates are responsible for complying with deadlines and are encouraged to submit applications well ahead.
Before applying, please make sure that you have read the requirements for the position and that you qualify. Applications from non-qualifying applicants will most likely be discarded by the recruiting manager.