Fraya — Values and Quality Standards

Fraya is a multi-skilled AI agent for e-learning content generation. This document describes what Fraya stands for — the quality principles that shape every design decision, prompt, and pipeline step.

These values are not aspirational slogans. They are the criteria by which every piece of generated content should be evaluated.

Multilingualism

Fraya creates native content for each language — not translations of an English original. A course in Dutch should read, feel, and reference culture as if it was written by a Dutch speaker. A course in Polish should do the same for Polish learners.

Supported languages:

This means: examples should be locally relevant, names should feel natural, idioms should not be literal translations, and cultural context should be appropriate for the audience. The localization asset exists specifically to enforce this across all sections of a course.

Accuracy

Fraya-generated content must be factually correct and free of fabrication. This includes: no invented statistics, no fake studies, no pseudo-frameworks presented as established theory, and no plausible-sounding but unverifiable claims.

Where Fraya cites a source — the source must be real and attributable. Where Fraya presents a framework — it must be grounded in established research.

This is a direction and a commitment, not a fully solved problem. Fraya is actively designed to reduce hallucination risk through the Knowledge Depository pipeline (which pre-loads verified source material before generation begins) and through blueprint-level pre-definition of factual anchors. Full accuracy verification is an ongoing development goal.

Learning motivation

Fraya-generated courses are designed to make learners want to continue, not just complete. This is achieved through three levers:

Narrative — every course is built around a central character whose story unfolds across modules. Learners follow a person like themselves through situations they recognise. Abstract concepts land differently when they're attached to a real situation.

Multimodal content — courses combine video, text, images, and interactive formats rather than delivering knowledge in a single mode. Each section type serves a specific cognitive purpose. Cognitive load is distributed deliberately — not all theory at once, not all exercises at once.

Digestible flow — course structure follows Bloom's progression: from recognition to application to creation. Learners are not overwhelmed early. Each module opens with a preview and closes with consolidation. Transitions between topics explain why the next step follows from the last.

Practical value

Fraya produces content useful in real work, not just in theory. The knowledge a learner gains should be applicable the next day — not just comprehensible in the moment.

This means:

• Examples are grounded in the learner's actual work context and challenges
• Action plans and practice tasks are included alongside conceptual content
• Assessments test application and judgment, not just recall
• The course concept is chosen based on what the learner needs to do differently, not just know better — pain and transformation are the starting inputs

Interactive formats (artifacts, scenario-based assessments) are a key part of practical value: they simulate real decisions rather than asking learners to describe what they would do.

Consistency

Every course Fraya generates must read as a single coherent work, not as a collection of independently generated sections. This is a structural challenge: section content is generated in parallel, by independent model instances with no shared context.

Fraya solves this through the asset pipeline:

• Characters asset — all section authors receive the same character descriptions, visual references, and arc summaries
• Visual bible asset — shared visual style reference for illustrations and video
• Localization asset — shared terminology and cultural decisions for all translators
• Injections — shared structural and naming rules applied uniformly across prompts

Consistency is also enforced at the outline level: blueprints pre-define factual anchors, character situations, and scope boundaries for each section before generation begins.

Personalization

Fraya is audience-first. Every course starts with a detailed learner profile — not a generic persona, but a specific description of who this person is, what they struggle with, what they already know, and what would make this course feel relevant to them.

This profile drives everything downstream:

• The concept is chosen around the learner's pain and desired transformation
• The narrative character reflects the learner's situation
• Examples are drawn from the learner's professional context
• Complexity level (Beginner / Intermediate) determines how deep the framework coverage goes
• Assessments match what the learner is actually expected to do in their role

Personalization is not about adding the learner's name to the content. It is about designing a course that could only have been made for this audience — and would feel generic to everyone else.