Knowledge Studio — Turning Existing Knowledge Into Active Learning

The problem that kept repeating

The inspiration for Knowledge Studio came directly from my work as a PhD student, where I give lectures and supervise students.

Whenever new students join a project, the same questions appear:

• What is Docker, and how do I use it?
• How do I configure the university server?
• How does a chatbot actually work?
• How is this research project structured?
• Where can I find the knowledge created by previous students?

The frustrating part is that these questions have often already been answered. Previous students have written bachelor’s or master’s theses, created presentations, documented projects, produced research papers, and implemented working prototypes. However, once a project is finished, much of that knowledge becomes passive archive material.

New students are then expected to read large collections of documents and reconstruct the relevant knowledge themselves.

The same problem exists outside universities. Companies invest significant time in onboarding new employees, even though much of the required knowledge already exists in internal manuals, presentations, videos, technical documentation, and previous project reports.

Knowledge Studio is based on a simple idea:

[ \text{Existing knowledge} + \text{specialized agents} + \text{human judgment} \rightarrow \text{interactive learning} ]

Instead of allowing valuable knowledge to disappear into folders, Knowledge Studio transforms it into something the next person can actively explore.

From archived material to interactive learning

Knowledge Studio converts existing source material into short, web-based, interactive learning experiences.

A user uploads documents such as:

• Bachelor’s or master’s theses
• Lecture slides
• Research papers
• Internal manuals
• Project reports
• Spreadsheets and structured data
• Images, audio, and video
• Source code and Jupyter notebooks

The application extracts and verifies the available knowledge, identifies the most important concepts, creates learning objectives, designs a suitable learning journey, and generates an interactive web experience.

The resulting learning nugget can contain:

• Structured explanations
• Quizzes and knowledge checks
• Matching and ordering exercises
• Drag-and-drop activities
• Reflection questions
• Flip cards and accordions
• Interactive diagrams
• Coding exercises
• Jupyter Notebook integrations
• Glossaries and key takeaways
• References to the original source material

The goal is not simply to summarize documents. Knowledge Studio turns passive information into an active learning process that helps users understand, apply, and retain knowledge.

Building a team of agents — not a black box

One of the most important design decisions was to avoid building one enormous prompt that receives all the files and produces a final website in a single step.

Knowledge Studio instead uses a pipeline of specialized agents. Each agent has a clearly defined responsibility, receives structured input, and produces a validated output for the next stage.

The pipeline includes:

1. Source extraction
   Documents, presentations, spreadsheets, media, code, and notebooks are converted into a common structured representation.

2. Content correctness checking
   Objectively verifiable claims are identified and checked. Potential problems are connected to their source location and presented for review.

3. Content analysis
   The verified material is transformed into concepts, relationships, learning objectives, content gaps, and pedagogical priorities.

4. Diagram planning and generation
   When a visual explanation would improve understanding, Knowledge Studio can create a diagram proposal and generate the corresponding image.

5. Pedagogical design
   The approved content map becomes a structured learning journey containing sections, interactions, navigation, assessments, and source coverage.

6. Learning-content generation
   Specialized agents create the explanations, quizzes, activities, glossary entries, and interactive elements.

7. Web experience building
   The approved content is assembled into a portable, responsive web experience.

8. Quality review
   A final reviewer checks factual grounding, pedagogical quality, accessibility, completeness, and technical correctness. If something needs improvement, only the affected part of the pipeline is repeated.

The agents exchange typed, structured artifacts instead of loosely connected text responses. This makes the pipeline more reliable and allows individual stages to be reviewed, edited, or repeated without restarting the entire process.

Human judgment at the center

The most important thing I learned was that human involvement should not be treated as an optional final check.

Knowledge Studio is designed as a collaboration between humans and AI agents.

The application pauses at important decision points:

• When factual problems have been detected
• After the content map has been created
• After the learning journey has been designed
• After the interactive content has been generated
• Before a generated diagram becomes part of the result

At each stage, the user can inspect the proposed output, make direct changes, request a focused revision, or approve the result.

This means the final learning experience is not simply the output of an invisible pipeline running in the background. It is the result of a transparent process in which agents perform specialized work while humans remain responsible for important decisions.

Designing for learning — not only content generation

A learning experience needs more than accurate text.

Knowledge Studio incorporates Bloom’s taxonomy to guide the learning progression:

1. Remember
2. Understand
3. Apply
4. Analyze
5. Evaluate
6. Create

The agents use this framework when deriving learning objectives, selecting interactions, designing assessments, and organizing the final experience.

For example, a learner might first receive a short explanation, then answer a knowledge check, apply the concept in an interactive exercise, and finally reflect on how it could be used in a real project.

This approach helps prevent the generated result from becoming another long document that the learner reads passively.

The hardest parts

Making multiple agents work together

The biggest technical challenge was defining how the agents should communicate.

Each agent needed:

• A clearly bounded responsibility
• A predictable input contract
• A validated output structure
• Access to the correct upstream information
• A defined route for revisions and failures

Without these boundaries, errors from one stage could propagate through the entire pipeline.

Balancing creativity and consistency

Interactive learning benefits from creative scenarios and varied activities, but unrestricted generation can produce experiences that feel disconnected or pedagogically weak.

The challenge was to support quizzes, diagrams, coding tasks, notebooks, interactive activities, and different learning structures while keeping everything aligned with the approved objectives and source material.

Keeping the result grounded in the sources

The generated experience must remain connected to the uploaded material. Knowledge Studio therefore preserves source provenance throughout the pipeline and verifies relevant factual claims before they are used downstream.

This was essential because a visually impressive learning experience is not useful if its content is unreliable.

Creating meaningful human review

Adding an approval button is easy. Creating review steps that show the right information at the right moment is much harder.

The interface needed to make evidence, concepts, relationships, learning objectives, activities, and generated content understandable without overwhelming the reviewer with internal technical details.

Processing very different source formats

Documents, slides, spreadsheets, images, videos, source code, and notebooks contain knowledge in very different forms. Converting them into a shared structure while preserving useful context was another major challenge.

What I am most proud of

I am proud that Knowledge Studio is more than a document summarizer or a single-prompt content generator.

It provides:

• A complete source-to-learning workflow
• Specialized agents with structured handoffs
• Human approval at consequential stages
• Fact-checking before downstream generation
• Source provenance in the generated experience
• Learning design based on Bloom’s taxonomy
• Generated diagrams and visual content
• Interactive activities and coding scenarios
• Portable web-based outputs
• A quality-review loop that can reroute only the work that needs improvement

I am also proud that the initial interface remains simple. Users only need to provide their source material, a project title, and the desired output language. The complex analysis and learning design happen inside a transparent, reviewable workflow.

What this project taught me

This project taught me that the quality of an agentic system depends less on the number of agents than on the quality of the connections between them.

Clear responsibilities, structured outputs, validation, and explicit transitions are essential.

I also learned that human feedback becomes most valuable when it is placed inside the process rather than added at the end. A reviewer should be able to influence the content map, learning journey, interactions, and factual corrections before those decisions become embedded in the final output.

Finally, I learned that educational content generation is not only a writing problem. It requires source analysis, verification, instructional design, interaction design, quality evaluation, and an understanding of how people learn.

Where Knowledge Studio goes next

The next step is to develop Knowledge Studio into a collaborative knowledge and onboarding platform.

Future improvements could include:

• Shared organizational knowledge libraries
• Reusable learning templates
• More advanced coding sandboxes
• Deeper Jupyter Notebook activities
• Additional interaction types
• Collaborative review and approval
• Learner progress and feedback analytics
• Personalized learning paths
• LMS exports using standards such as SCORM or xAPI
• More languages and accessibility options
• Durable multi-user projects with authentication and role management

My long-term goal is to make existing knowledge easier to preserve, reuse, and teach.

Universities, research groups, schools, and companies already possess enormous amounts of valuable material. Knowledge Studio helps transform that material from something people are expected to read into something they can actively experience, understand, and apply.

Built With

• css
• fastapi
• html
• javascript
• openai
• postgresql
• python