Description

The LLM project aims to integrate artificial intelligence in the form of Large Language Models (LLMs) into the XWiki platform. This project, led by Ludovic Dubost and Matéo Munoz, an intern at XWiki, seeks to leverage AI capabilities to enhance the functionality and user experience of XWiki. The planned features of the project are listed there.

One of the primary objectives is to do some classical generation tasks (create content based on a prompt, or summarize content) using multiple LLM options to users within XWiki, allowing them to choose their preferred model based on personal preference, either ChatGPT but also Open Source models that can be executed locally without any privacy risks. Additionally, the project is designed to be modular, enabling easy future enhancements and advancements, including using AI to search for Wiki content in natural language, or to also allow to perform tasks on the Wiki using instructions in natural language.

For a detailed understanding of the project's architecture, please refer to the provided page.

Initially, we will prototype the first features using OpenAI ChatGPT (utilizing GPT-4). Once the main features demonstrate satisfactory performance, the development process will expand to incorporate Open Source LLMs and deliver a well integrated UI in the XWiki UI.

By June 30, 2023, a preliminary prototype will be launched, offering users a glimpse of the upcoming UI design. This prototype will include the core functionalities and showcase the integration of multiple LLM options.

You can stay updated on the progress of the project by referring to the provided project plan, which will be continuously updated to reflect the current status of the project.

Key Documents

• Features
• Task schedule
• Generative and Summarization Components Architecture

Research documents

Ideas and Technological research for this project

• Brainstorming about AI In XWiki
• AI Ideas
• Wiki AI Search Engine (WAISE)
• Competitive Analysis
• Performance and Costs Measurements
• AI API in use
• GPT based Technologies