Learning in communities of practice (CoPs) makes up a significant portion of today's knowledge gain. However, only little technological support is tailored specifically toward CoPs and their particular strengths and challenges. Even worse, CoPs often do not possess the resources to host or develop a software ecosystem to support their activities. In this contribution, we describe a decentralized learning infrastructure for community knowledge building. It takes into account the constant change of these communities by providing a lightweight and scalable infrastructure, without the need for central coordination or facilitation. As a real use case, we implement a question-based dialog application for inquiry-based learning and ignorance modeling with our infrastructure. Additionally, we explore the possibility of using social bots to connect the services provided by the decentralized infrastructure to communication tools already present in most communities (e.g., chat platforms). Following a design science approach, we describe a multistep evaluation of both the infrastructure and application, together with the improvements made to the resulting artifacts of each step. Our results indicate the relevance of our approach, which may serve as an example of how decentralized learning infrastructures for learning outside of formal settings can be applied by CoPs for knowledge building.