Welcome To The World Of Decentralised Social Media: XWiki

Decantralised Social Media - XWiki

XWiki

Michael Hamann
Michael Hamann – R&D Engineer at XWiki

In our daily lives, whether personal or professional, capturing, organizing, and retrieving information is essential. Wikis, particularly XWiki, are a great way to organize information and foster collaboration. However, with the flood of information we’re dealing with, it can be overwhelming to find what we need. That’s where Large Language Models (LLMs) come in — they have an amazing capability to “understand” human language, sift through huge amounts of text, summarize information, and adapt texts to different use cases, to name just a few of the capabilities of which we discover more every day. While there are justified ethical concerns about the use of LLMs, I believe LLMs can be a valuable tool.

 

Our work on the WAISE project, funded by NGI Search, has allowed us to take the first step in integrating LLMs into XWiki, making it easier to manage and access information.


The What

WAISE (Wiki AI Search Engine) is a conversational AI search server developed by XWiki as part of the NGI Search initiative. It employs Retrieval-Augmented Generation (RAG) to combine retrieved content with AI-generated insights, delivering context-aware responses in natural language. Built on XWiki’s extensible platform, WAISE integrates with other CMS systems and external tools, offering secure and efficient knowledge management capabilities such as querying, content summarization, and transformation. Its design ensures compliance with user access rights, maintaining both flexibility and data privacy.

 

For What

WAISE was created to address the growing challenges of navigating large and complex knowledge bases, both within XWiki and beyond. As organizations rely on ever-expanding data repositories, finding relevant, actionable information can become a bottleneck. By integrating WAISE, XWiki users gain access to AI-enhanced tools that simplify workflows, provide instant answers to complex queries, and support multilingual environments. The project also aligns with XWiki’s vision of fostering open-source, ethical innovation by allowing to use both cloud based LLM services as self-hostable open weight and open source models.

 

THE REACH

WAISE is designed to benefit organizations of all sizes and across diverse sectors, including technology, education, healthcare, and public administration. Its compatibility with self- and cloud-hosted LLMs allows for adaptability to specific workflows and industry needs. Companies focused on privacy, secure collaboration, or multilingual knowledge management can leverage WAISE to enhance their operations. Additionally, its integration with the Matrix protocol brings conversational AI into real-time chat environments, making it particularly valuable for distributed teams and global communication.

 

THE MILESTONES

The NGI-funded WAISE project has made significant advancements in enhancing AI-powered search and collaboration. Key milestones include the integration of embedding-based semantic search, keyword search, and hybrid search techniques, which have significantly improved the accuracy and efficiency of information retrieval. The creation of a Matrix bot enabled real-time conversational AI linked to XWiki content, providing seamless integration into collaborative workflows.

Extensive benchmarking of open-source LLMs offered critical insights into their performance, guiding improvements in tasks such as question answering and summarization. Additionally, WAISE introduced a REST API for indexing content from external applications, enabling expanded functionality beyond XWiki. Secure authentication mechanisms were implemented to maintain user access controls during AI interactions. Moving forward, efforts are focused on enhancing usability, particularly simplifying the configuration UI to ensure WAISE is accessible to technical and non-technical users alike.

XWiki

Publication Date

  16/01/2025

CATEGORY

OTHER PROJECTS

Skip to content