Agenda

At Ubisoft, knowledge is scattered across Confluence, Jira, GitLab, video platforms, and dozens of team silos. This fragmentation makes it harder to ramp up, debug, plan, or even stay aligned. So we built DocGPT: an internal, production-grade assistant designed to help teams search, understand, and reuse Ubisoft's knowledge using natural language.
Our goal was simple: reduce context switching and make internal knowledge more accessible. We co-developed our tech with production teams. Then it scaled. Fast!
Today, DocGPT supports workflows across multiple Ubisoft productions. It helps testers generate test cases from Design docs. It assists new developers with step-by-step guidance or by explaining custom pipeline code. It even supports managers by surfacing project status and Jira insights in plain English.
In this talk we'll share honest lessons from our rollout — from organizational acrobatics to surprising wins that reshaped our roadmap.

Takeaway

In this talk, you'll learn:
• How we built a secure, modular RAG architecture to serve real-time answers from game production data
• How we tackled challenges around proprietary knowledge , assistant customization, and knowledge freshness
• Take away a practical roadmap to help you build and deploy your own GenAI assistant in a production environment
• What worked, what broke, and how we adapted the mindset around GenAI to address long-endured pain points

Intended Audience

This talk is for producers, technical directors and other game leaders who are interested in reducing time spent in searching for information to reinvest their time in building games and shipping faster. No prior machine learning knowledge is required, but familiarity with AI/ML concepts is helpful.