Volunteer AI Systems Architect Public Impact

African Development Choices (ADC) is a UK-registered charity on a mission to reduce poverty by empowering communities to make more informed choices in the management of public resources.

ADC’s purpose is to enable communities to independently meet their needs for essential services, such as clean water and safe sanitation, and improve lives in Africa.

At ADC we encourage a working culture of seeking and giving input, showing and taking initiative, and commitment to the organisation and its purpose.

We now have an exciting opportunity for a volunteer Senior AI Agent Architect to join ADC and help build a live prototype of Maji Guard AI.

To lead the technical design and hands‑on development of Maji Guard AI, an agentic AI prototype that helps ADC identify underused WASH resources, structure recovery and reallocation cases, and verify that recovered value reaches facility upgrades and operations and maintenance.

Technology / AI Innovation

Product & Engineering

Head of AI Innovation

Volunteer AI engineers, developers and technical contributors

• Lead the end‑to‑end architecture of Maji Guard AI, including the selection and integration of models, agent workflows, orchestration tools, retrieval systems, data structures and human‑in‑the‑loop controls appropriate for a live prototype.
• Translate ADC’s operational goals into practical AI workflows, especially for identifying underused WASH resources, standardising evidence and case preparation, supporting recovery and reallocation decisions, and verifying delivery of value to facility improvements and ongoing operations and maintenance.
• Build and iterate a live prototype using pragmatic, cost‑effective tooling, with strong attention to reliability, traceability, security, maintainability and ease of future scaling.
• Design structured inputs, prompts, evaluation criteria and decision logic so that outputs are useful to non‑technical ADC teams working across programmes, research, partnerships and leadership.
• Establish a technical roadmap for the prototype, including architecture choices, milestone planning, testing priorities, deployment options and documentation for future funders, partners and developers.
• Work closely with ADC leadership to shape the product vision and ensure the prototype aligns with ADC’s wider governance and public service innovation model.
• Advise on suitable approaches for data governance, privacy, model risk, hallucination control, auditability and responsible AI use in the context of public‑resource management and service delivery.
• Define and monitor prototype success metrics, including usefulness, accuracy, turnaround time, workflow fit, and operational learning that can strengthen grant applications, partnerships and future product development.
• Support the transition from local experimentation to a live prototype environment, including basic cloud or hosted deployment thinking, version control practices and engineering handover readiness.
• Mentor and coordinate other volunteer technical contributors where relevant, helping ADC build a small but high‑quality AI capability around Maji Guard AI.

• 8+ years of relevant experience in AI, machine learning, applied NLP, intelligent automation, agentic systems, or modern software engineering for production‑grade data or AI products
• Strong hands‑on experience designing and building AI systems, including LLM‑based applications, retrieval‑augmented generation, agent workflows, orchestration frameworks, evaluation pipelines and API integrations
• Experience taking ideas from concept or prototype stage into a usable live product, pilot or production‑like environment
• Excellent technical judgement on model selection, system design trade‑offs, latency, reliability, observability and cost‑conscious engineering
• Strong programming ability, ideally including Python and modern backend tooling, with the ability to work across architecture, implementation and troubleshooting
• Experience designing structured workflows where AI outputs must be reviewable, auditable…