chunkx is an AI-first corporate learning platform that generates learning content individually for each employee and adapts it to their learning needs. This approach combines hyper-personalized learning with a framework that companies can shape: they set the learning objectives, subject-matter sources and requirements. Within this framework, the AI develops fitting explanations and exercises. Ongoing quality checks and human feedback help to detect deviations and improve the application.
How much freedom does such an AI need, and how do companies stay in control? I discussed this question from our implementation perspective at the Handelsblatt conference AI in Insurance 2026 in Düsseldorf on September 16 and 17. In regulated fields of work in particular, the task becomes concrete: an explanation should fit the person, while the underlying subject-matter rule must stay intact.
What AI First Means in Corporate Learning
At chunkx, AI first means that AI helps shape each individual's learning process. The platform generates microlearning units based on company knowledge and the context of each course. Prior knowledge, role, learning needs and feedback influence how content is taught. Personalization covers the content itself, how it is explained and the exercises that go with it.
An example scenario: two employees are learning about the same insurance benefit. A new colleague first needs the terms explained, followed by a simple use case. An experienced colleague wants to understand an exception and practice on a more demanding case. Both work from the same subject-matter foundation. What helps each of them understand it differs.
We call this approach hyper-personalized learning. It combines the selection of relevant topics with the individual generation and adaptation of learning content. Learners can ask follow-up questions, have an explanation simplified or ask for an example from their own work context. We describe the step from selection to generation in more detail in our article on adaptive content creation.
The result is an ongoing learning interaction: employees get support, practice applying what they learn and give feedback when something is still unclear. Subject-matter owners shape the foundation for this. They define goals and limits and check whether the individual teaching stays within them.
What Control Companies Have over the AI
A good trainer responds to difficulties in understanding, finds other explanations and asks the right follow-up questions. At the same time, a good trainer knows which subject-matter rules must not be changed. We design the AI's latitude on the same principle: the teaching may be individual, while the subject-matter requirements remain binding.
In the chunkx Creator, those responsible define the course context, learning objectives, mandatory topics and access rights. They upload documents and add instructions for how the content is generated. Source grounding helps determine how closely the AI should stick to the documents provided. This way, the framework can be shaped to fit the purpose of the learning.
| Learning situation | Individual latitude | Binding framework |
|---|---|---|
| Product knowledge | Language, depth of explanation and fitting examples | Product features, conditions and exclusions |
| Role play | Course of the conversation, follow-up questions and reactions | Learning assignment, subject-matter rules and behavioral limits |
| Regulated training | Learning path and additional practice | Required content and defined evidence |
This distinction is crucial for companies. A tight subject-matter framework still leaves plenty of room for personal support. At the same time, the sources and instructions themselves have to be right. Contradictory or outdated requirements cannot be resolved by more personalization alone.
How We Check the Quality of Generated Learning Content
Source grounding and clear instructions already take effect during generation. However, they do not eliminate every possible deviation. That is why checking in live operation is part of our approach. Our quality process combines automated checks of generated learning units with user feedback and subject-matter assessment by people. Insights from operation feed into further development.
A real case shows why this matters: in a generated learning unit, the learning text explained a subject-matter rule correctly. The model answer to the quiz question that followed, however, deviated from it. A subject-matter reviewer reported the inconsistency. Learners could have received contradictory feedback despite understanding the rule correctly.
We investigated the case and developed our quality assurance further. The key insight: learning text, task and model answer must also be consistent with one another. What counts for learning quality is whether the unit as a whole conveys a coherent understanding.
Cases like this help us improve our checks in a targeted way. The yardstick is the specific purpose of the learning: Is the subject-matter statement preserved? Does the task actually test the knowledge that was taught? And does the feedback help the learner understand a mistake?
Why an AI That Checks Quality Must Be Checked Too
Automated quality checks can be wrong as well. In one of our real cases, content that was acceptable on the subject matter was flagged unnecessarily. We adjusted the check as a result. The example shows why, at chunkx, we also check the quality of our checking procedures.
What matters for companies is that a check catches actual errors while accepting helpful, acceptable content. Unnecessary flags cost subject-matter owners time and can restrict useful personalization. Both sides belong to our understanding of quality.
How Documented Limits Support Use in the Company
Besides settings in the product, companies need a clear description of its intended purpose. To assess our use of AI in the context of the EU AI Act, we have prepared three related documents: the Intended Purpose & Classification Statement describes the intended use and how we classify the system. The Provider Documentation explains functions, architecture, data processing and safeguards. The Deployer Guideline translates this framework into requirements for use at the customer.
One key limit concerns how learning feedback is used. At chunkx, individual AI feedback is intended to support learning. It should not become the basis for personnel decisions or formal qualification decisions. If such a use is planned, the deployment must be reassessed. That is why we also consider who receives which information and what reports are used for.
This documentation gives the departments involved a common basis for their review and for decisions about new use cases. We give an overview on our security and compliance page.
Where the Value of Hyper-Personalized Learning Should Show
For L&D leaders, what counts is whether people understand the knowledge they need and can apply it in their day-to-day work. That is why, besides learning activity and completions, companies should also look at fitting application situations: Can employees transfer a product rule to a new case? Do they need less support with recurring tasks? How much subject-matter review work remains for those responsible?
Questions like these can be examined in a pilot project with a baseline measurement and repeated tasks. Our aim with chunkx is to make individual learning support available to many people. Companies set the subject-matter direction, while the AI opens up different paths to understanding. The value has to show in helpful learning experiences and verifiable quality.
Want to see how this approach works with your company's knowledge? In a chunkx demo, we'll discuss your learning use case, the right amount of latitude and your quality requirements.
