Mongolia is introducing the Teacher Virtual Assistant through Medle, its national learning platform, in a phased manner. The tool will help teachers generate lesson plans and differentiated practice questions for Grades 1–5 mathematics. The larger lesson from Mongolia is not about one AI application, but about the system being built around it: stronger education data systems, government-owned, curriculum-linked digital content, school quality standards, and early work on sustainable financing.

Governments around the world are rapidly piloting AI-enabled education tools, ranging from lesson-generation models and adaptive tutoring systems to teacher-facing assistants. Yet many promising pilots do not progress to system-wide adoption. The bottleneck is often not the technology itself, but whether education systems are ready to absorb, govern, and sustain AI at scale. Student records, attendance data, learning platforms, teacher systems, and curriculum repositories are often developed separately. Financing is often project-based, while responsibility for digital systems is spread across multiple agencies.

Drawing on operational work across education systems, the World Bank Education and Digital teams developed a three-layer architecture for moving AI in education from pilots to scale. This post introduces the framework and shows how Mongolia is putting it into practice.

A Three-Layer Architecture for Scaling AI

The architecture consists of three connected layers: Layer 1 establishes trusted data and institutional foundations; Layer 2 uses those foundations to deploy reusable AI applications; and Layer 3 supports scale through financing, operations, and partnerships.

Layer 1: Digital Public Infrastructure for Education

The first layer is the foundation. AI tools can only provide reliable personalization if education systems can identify learners consistently, link them to curriculum and assessment data, and update records across platforms. This requires unique student and teacher identifiers, interoperable education information systems, school connectivity, assessment and attendance systems, and clear governance arrangements.

Under the Global Partnership for Education-financed System Capacity Grant, Mongolia has strengthened the foundations of its education data systems. This includes introducing cross-system student identifiers, improving interoperability across education databases, and strengthening data quality and validation rules. In parallel, the country developed its first Education Quality Standards Framework, covering digital readiness, infrastructure, learning materials, and teacher capacity. These reforms may appear technical, but together they create the conditions for AI-enabled education systems to be built.

Layer 2: Modular AI Architecture and Platforms

The second layer is where AI applications are deployed. Rather than building separate systems for each new tool, governments can use modular AI platforms built around reusable components, such as curriculum repositories, learner profiles, assessment tools, analytics, and data exchange services. This reduces duplication, limits vendor lock-in, keeps systems aligned with the national curriculum, and allows technology to evolve without rebuilding the underlying platform.

The Teacher Virtual Assistant is Mongolia’s first application of this modular layer. Delivered through Medle, it helps teachers generate lesson plans and differentiated practice questions for Grades 1–5 mathematics, drawing on a government-owned curriculum dataset comprising more than 700 lesson plans and 4,900 practice questions.

The significance of the TVA extends beyond the tool itself. It demonstrates how AI applications can be integrated into a government-owned platform rather than operating as stand-alone products. Because it is built on national systems and curriculum content, it provides a pathway for introducing additional tools in the future, including adaptive assessment, content recommendation, and teacher feedback. The phased rollout is expected to generate evidence on teacher adoption and how AI-enabled lesson planning may support classroom practice and learning over time.

Layer 3: Financing and Private Sector Mobilization

The third layer focuses on scale. Many AI pilots are funded through short-term projects and are never integrated into regular government systems. Sustained deployment requires financing models that can support infrastructure, maintenance, teacher support, and future upgrades. It also requires clear roles for governments, development partners, and the private sector.

In Mongolia, this layer is being explored through education PPP discussions supported by the World Bank Group, including IFC and MIGA. The immediate focus is on whether private participation can help address infrastructure constraints, including classroom shortages in overcrowded urban schools, while keeping public oversight over quality and equity. This matters for the AI-readiness agenda because digital and AI reforms cannot scale in isolation. They require schools with adequate space, connectivity, basic infrastructure, and sustainable financing arrangements.

An important lesson from Mongolia’s experience is that the financing discussion was strengthened because the first two layers were already underway. The data systems strengthened under the System Capacity Grant, the quality benchmarks established through the Education Quality Standards Framework, and the modular logic demonstrated through the Teacher Virtual Assistant provided a concrete foundation on which longer-term financing options could be explored.

For other countries, Mongolia’s experience shows that the lesson is not to wait for perfect systems before using AI, but to ensure that each AI application strengthens the systems needed for the next one. AI will not scale through tools alone. It will scale where countries build the data systems, platforms, institutions, and financing arrangements needed to absorb and sustain innovation over time.

By Salman Asim, Hayeon Kim & Alex Twinomugisha

Published Date:2026-07-22