Project summary

COMPASS is the first-of-its-kind AI governance fellowship built specifically for active civil servants in the Global South, people who already have institutional power and can influence AI policy from day one of the program.

Artificial intelligence is reshaping how governments work. From automated public service delivery to algorithmic border controls, from national AI strategies to international treaty negotiations, the decisions being made right now will define governance norms for decades. The problem is that the people making those decisions in the Global South are largely doing so without structured training, mentorship, or peer networks.

Most existing AI governance fellowships target students, researchers, or private sector professionals. They are excellent programs. But they share a critical limitation: their graduates still need to find jobs, build credibility, and navigate bureaucracies before they can actually move policy. That process takes years, years the world does not have.

COMPASS takes a fundamentally different approach. It targets people who are already inside government, civil servants, regulators, and public sector officials across Africa, South Asia, Southeast Asia, and Latin America, and gives them the technical fluency, governance frameworks, and peer network they need to act immediately on what they learn.

What are this project's goals? How will you achieve them?

The goal is to close the gap between people who already have the power to shape AI policy and the technical grounding they need to use that power well. Most AI governance training right now goes to students and researchers who are years away from actually influencing anything. Meanwhile the civil servants writing procurement contracts, advising ministers, and sitting in UN negotiations today get almost no structured support. COMPASS is built to fix that mismatch by working directly with people who can act on what they learn right away, not eventually.

By the end of the fellowship, each fellow should have:

• practical literacy in how AI systems work, where they break, and what that means for public institutions (not deep technical training, just enough to ask the right questions and spot bad vendor claims)

• working knowledge of the major governance frameworks (OECD, UNESCO, the EU AI Act, the AU Continental Strategy) and, more usefully, what those frameworks actually require of someone running a procurement process or drafting legislation

• a peer network of counterparts across Africa, South Asia, Southeast Asia and Latin America facing the same institutional constraints

• one real deliverable: a Governance Action Memo applying what they've learned to an actual problem in their own institution

Additionally, we're aiming for a working alumni network, around 150 trained officials across five cohorts, some of whom go on to shape national AI strategy or represent their countries more effectively in multilateral rooms.

Here's how we get there. Ten weeks, fully online, two cohorts of 30 fellows in year one. Each week pairs self-paced reading (3-4 hours, timed to fit around a government job) with a live 4-5 hour weekend session run by an actual practitioner, someone who has sat in the room, not just written about it. The curriculum is built around Global South realities from the start, bureaucracy, thin resources, vendors from competing geopolitical blocs, rather than adapted from EU or US frameworks after the fact. Selection prioritizes people already inside government with real seniority (5-15 years in, working on tech policy, AI adoption or procurement), with extra weight for officials from countries actively building AI strategy right now and for fragile or conflict-affected states where governance capacity is thinnest.

How will this funding be used?

The $84,700 covers year one in full: two cohorts, 60 fellows, roughly $1,400 per fellow.

Most of it goes to people: a part-time Program Director ($36,000) handling curriculum design, mentor coordination and quality control, a part-time Program Coordinator ($12,000) running applications and day-to-day fellow support, and a small stipend ($3,000) for one or two volunteers.

The next biggest chunk goes to actually running the program: building out curriculum and session materials ($9,000), honoraria for practitioner mentors at $500 per session across 10 mentors per cohort ($10,000), and a modest budget for guest speakers on specialist topics ($2,500).

The remainder covers the basics: video conferencing and LMS platform costs ($1,800), outreach and the application process including translation ($1,500), and monitoring and evaluation so we can actually report back on what worked ($1,200), plus a 10% contingency ($7,700) for the stuff you can't plan for.

We expect the per-fellow cost to drop 30-40% in year two once curriculum and mentor relationships are already built, with year three introducing a modest certification fee for employer-sponsored cohorts to start reducing grant dependency.

Who is on your team? What's your track record on similar projects?

Techplomacy Foundation is an independent, non-partisan nonprofit working at the intersection of tech, AI, diplomacy and governance. We publish Techplomacy Magazine, covering AI and technology governance with a specific focus on the Global South, so this isn't a new subject area for us, it's the thing we already spend our time on.

Our core team and advisors include:

Olin Thakur, Co-Founder (Strategic Operations & Change Leadership, AI Governance, Techplomacy.
Leveraging 18 years’ experience across 21 countries).

Danny Padilla, Former UNESCO Chief of Education

A. Rogers, Former UNDP HQ Communications Advisor

and other former senior UN officials

What are the most likely causes and outcomes if this project fails?

Recruitment and mentors are the two risks we're watching. That's why the program is part-time and self-paced (6-8 hours a week), built around a sitting civil servant's schedule, not a student's. And we're lining up mentor relationships in advance rather than scrambling once a cohort starts.

Even in a good cohort, some fellows will hit bureaucratic walls back home that training alone can't fix. That's not a program failure, it's why week 9 exists.

Worst case, we get a smaller or slower first cohort, not a failed one, since year one is funded regardless. The real risk is longer-term: proving the model well enough to land employer-sponsored seats down the line. Getting year one right is what that depends on.