The Kelias Methodology: How the Curriculum Was Built and Why All Three Sources Matter
Most AI literacy curricula are built from the top down. An expert or team defines what AI literacy means and builds content to explain it. Kelias was built from three converging sources simultaneously:
- a regulatory text (Article 4 of the EU AI Act),
- a competence framework (DigComp 3.0), and
- role-derived questions from Lithuanian public sector AI use contexts, built through role analysis rather than practitioner interviews.
Each source contributes something the others cannot. When any one is absent, the curriculum loses a quality that matters specifically for Article 4 compliance purposes.
Key Takeaways
- Source 1 (Article 4) provides the legal anchor: every module maps to an obligation the regulation actually creates
- Source 2 (DigComp 3.0) provides the competence structure: every module targets specific competences at specific proficiency levels, producing assessable and documentable outcomes
- Source 3 (practitioner questions) provides context specificity: the scenarios that make the legal and framework content meaningful to a Lithuanian case worker or IT lead
- All three sources are needed: a curriculum without the legal anchor lacks regulatory grounding; one without the framework lacks measurable outcomes; one without practitioner context lacks operational relevance
- The Kelias methodology is disclosed as systematic derivation at pilot stage, not as formally validated curriculum science; pilot feedback is the primary validation input
Source 1: Article 4 as the Legal Anchor
The curriculum design started with Article 4 of the EU AI Act, as amended by Regulation (EU) 2026/1744. The amended text sets out what Article 4 actually requires: measures to support AI literacy development, calibrated to role and context, considering the people affected by AI systems.
Working from the text, several questions emerge that the curriculum must equip learners to answer in practice:
What does it mean to use AI safely and with awareness of its limitations? (The "safe use" dimension of literacy.)
What does critical engagement with AI output require? (The "critical use" dimension: not accepting outputs uncritically.)
What are the governance obligations that apply when the institution deploys AI in decisions affecting citizens? (The "institutional context" dimension.)
What does meaningful human oversight look like in the specific AI systems each role uses? (The "context-specific oversight" dimension.)
Each of these dimensions maps to specific module content. The legal anchor ensures that the content is not just AI education in the abstract; it is AI literacy education grounded in what the regulation asks of the people completing it.
A curriculum built without the legal anchor produces AI awareness content. It may be accurate, useful, and well-designed. It cannot, however, be pointed directly at Article 4 obligations, because those obligations are not its organising principle. Completions do not constitute measures to support AI literacy development in the Article 4 sense; they constitute good digital education.
Source 2: DigComp 3.0 as the Competence Structure
DigComp 3.0 (JRC144121, November 2025) is the second structural source. Every module and every lesson in the Kelias curriculum maps to specific DigComp competences and proficiency levels. The learning design sequence is:
- identify the target DigComp competence and level for a specific role,
- determine what a learner needs to be able to do to demonstrate that competence at that level,
- and design content that develops and can assess that capability.
This approach produces two things that the Article 4 source alone cannot provide.
First, measurable learning outcomes. DigComp proficiency descriptors are specific enough to assess: a learner either can or cannot evaluate an AI output for an anomaly in a non-routine case context (Level 4) versus a familiar case context (Level 3). That specificity makes pre- and post-assessment possible and makes the assessment results meaningful.
Second, the competence record that Article 4 compliance requires. The competence report that Kelias produces per learner references specific DigComp competences and levels. It is externally verifiable, comparable across institutions, and legible to a supervisory authority reviewing Article 4 compliance documentation.
A curriculum built without the DigComp framework may be legally grounded (through the Article 4 source) and practically useful (through the practitioner source). What it cannot produce is a competence record that tells a regulator which specific competences a learner has, at what proficiency level. Without the framework, the outcome is AI literacy training. With the framework, the outcome is documented AI literacy competence.
The DigComp 3.0 framework is available at: https://publications.jrc.ec.europa.eu/repository/handle/JRC144121
Source 3: Role-Derived Context
The third source is the one that distinguishes Kelias content from content that is accurate but abstract. Role-derived questions are built from role analysis: for a Lithuanian case worker using an AI benefits scoring tool, what does that person actually encounter, what goes wrong, what decisions do they face, and what does AI literacy change about how they handle those situations?
These questions were developed through role-specific analysis of Lithuanian public sector AI use contexts, informed by the specific AI applications most common in Lithuanian public administration (benefits assessment, document processing, eligibility systems) and the specific Article 4 obligations most relevant to each role. They are not based on interview data from a validated study. This is a transparency disclosure; the role analysis is systematic and grounded in the actual Article 4 text and DigComp framework; it is in the process of being validated through the pilot programme.
Lithuania is the initial target market and the context used to build the first curriculum iteration. The underlying methodology is not Lithuania-specific. Article 4 applies across all 27 EU member states, and the role analysis approach — mapping roles to AI systems, then to DigComp competences, then to context-specific scenarios — applies equally to a French municipal officer, a German HR manager, or a Dutch benefits assessor. Future development will extend the role-derived source to additional member state contexts and private sector use cases, producing locally relevant scenario content for each. The methodology stays constant; the scenarios change.
What the role-derived source contributes is specificity that neither the legal text nor the framework can provide alone. Article 4 says to take literacy measures "in the context the AI systems are to be used in." DigComp says Area 5.1 competence at Level 3 means the learner can solve problems with AI in familiar contexts independently. Neither source tells you what a Lithuanian social services officer does when a risk assessment flags a stable family for intensive intervention. The role-derived source does.
Content that lacks this dimension is accurate and framework-aligned but teaches to a generic version of the role rather than the actual one. Learners who complete it know about AI literacy; they may not know how to exercise AI literacy in the specific moments their job creates. The gap between knowing and doing is where the role-derived source sits.
How the Three Sources Interact in a Module
Module 5 (Safety, GDPR, and Ethical AI Use) illustrates the three-source interaction.
The legal source identifies the Article 4 obligations that apply to this domain: awareness of AI risks, understanding of GDPR Article 22 and its implications for cases where AI informs decisions affecting citizens, and understanding of the institutional obligations under Annex III for high-risk AI systems. This defines what the module must cover.
The DigComp source maps the module to Areas 4.1 (protecting personal data), 4.2 (data protection in AI contexts), and 4.3 (protecting institutional wellbeing and understanding citizen rights). It sets the proficiency level targets: Level 3 for case workers, Level 4-5 for legal officers and IT leads who carry broader responsibility for these obligations. This defines what competence level the module should develop and how the assessment should be designed.
The practitioner source translates these into scenarios: a case worker whose AI tool processes personal health data as part of a social care eligibility assessment — what do they need to understand about what the system is doing with that data, what the citizen's rights are, and what they should do if the citizen asks about it? These scenarios make the legal-text obligations and the framework-competence statements concrete and testable.
The resulting module content is: legally grounded (Article 4 compliance rationale throughout), framework-structured (DigComp Area 4 competences, Level 3-5), and contextually relevant (Lithuanian public sector scenarios that the learner recognises from their own work).
The Module Map
Kelias organises content across seven modules, each designed for specific DigComp areas and proficiency levels:
Module 1 (What AI Is and Isn't): DigComp Area 1 at Foundation to Intermediate level. Foundational understanding of AI systems, their capabilities, and their limitations. Entry point for all learners.
Module 2 (AI in Data and Information Tasks): DigComp Area 1, deeper, and Area 2 introduction. Using AI tools for information tasks, critically evaluating AI-generated outputs.
Module 3 (AI in Communication and Collaboration): DigComp Area 2. How AI mediates communication in public sector contexts, citizen-facing AI interaction, managing AI-assisted collaboration.
Module 4 (AI-Generated Content): DigComp Area 3. Understanding, creating, and critically evaluating AI-generated content in a public sector context. Copyright and authenticity implications.
Module 5 (Safety, GDPR, and Ethical AI Use): DigComp Area 4. Data protection, GDPR Article 22, Annex III awareness, ethical dimensions of AI use in public administration.
Module 6 (Critical Evaluation and Override): DigComp Area 5. Problem-solving with AI, recognising when to override AI outputs, escalation processes, documentation of reasoning.
Module 7 (AI Governance and Strategy): DigComp Areas 4+5 at Advanced level (5-6). Institutional AI governance, Article 4 programme design, procurement governance, regulator engagement. For senior leaders and AI governance roles.
The Transparency Disclosure
The Kelias curriculum is built on systematic derivation from three documented sources. It has not been through formal academic curriculum validation or psychometric assessment validation. The pilot programme is the primary vehicle for validating both the curriculum design and the assessment tools.
Formal validation takes time and typically follows demonstration of practical value. What Kelias has done at this stage is build with method, from sources that are publicly documented and externally verifiable, and be transparent about the stage of development. Pilot learners and institutions know what they are engaging with.
Feedback from pilot learners is the validation input. Where the curriculum produces unexpected results (learners who struggle with content designed for their level, assessment questions that produce ambiguous results, scenarios that do not resonate with the actual work), the curriculum is updated. The transparency about method is how that feedback cycle stays honest.
Kelias is a free, open 6-month AI literacy platform for EU professionals using AI systems, mapped to DigComp 3.0 and built for Article 4 compliance. Access at kelias.tech or contact joseph@kelias.tech.
Ikpong Joseph Alexander holds an MSc in Artificial Intelligence and is the founder of Kelias. Sources: Regulation (EU) 2024/1689, Article 4 (as amended by Digital Omnibus, Regulation (EU) 2026/1744, July 2026); JRC144121 DigComp 3.0, Joint Research Centre, November 2025, available from the JRC publications repository. This post does not constitute legal advice.