Collecting knowledge fails because of people. So we built an AI colleague.

Fabian Hagen explains why traditional knowledge management systems often fail in organizations due to human inconsistency. Instead of forcing employees to manually document information, his team built 'Frieda,' an internal AI colleague that autonomously collects and processes data from Slack, Confluence, and GitHub. Frieda functions as a RAG-based assistant, providing accurate, source-backed answers while also automating code reviews. The system runs on internal infrastructure using Node.js, PostgreSQL, Qdrant, and FalkorDB, ensuring data remains where it belongs while making it accessible through a unified interface. By shifting the burden of documentation from the team to an automated, rule-based toolset, the company aims to maintain a living knowledge base without the overhead of manual curation. The project highlights the importance of building custom agentic workflows over relying on generic off-the-shelf solutions, emphasizing that the success of such systems depends heavily on team integration and well-defined technical pipelines.