Georgia-Pacific's ChatGP Combines Claude on Amazon Bedrock with Live IoT Data to Guide Machine Operators

Georgia-Pacific built an internal generative AI assistant called ChatGP that gives machine operators and maintenance technicians answers drawn from documents, maintenance records and live machine data, which the company says has reduced off-quality output, cut machine downtime and raised productivity. Vice president of applied AI and products Roshan Shah estimated potential savings of millions of dollars a year across its facilities, according to an AWS case study. "It's like having a digital expert who's always there, ready to help both new and seasoned operators find the right answers fast," said Ryan Holbird, senior manager of digital transformation at Georgia-Pacific. The company runs more than 140 facilities with varied equipment, and the know-how to maintain and tune it was scattered across binders, digital files and the memories of veteran staff. As experienced employees retired, junior operators lost access to that knowledge, leading to longer downtime and higher repair costs. Georgia-Pacific worked with AWS Professional Services and, starting in early 2023, used Amazon Bedrock to access Anthropic's Claude for understanding and answering questions. Amazon Kinesis streams Internet of Things sensor data into the system, so when an operator asks about a machine problem ChatGP can combine text guidance with the machine's current state, recent performance trends and likely causes. Operators use a web app on any computer or tablet on the company network, receive step-by-step guidance on inputs and adjustments, and get answers tailored to the equipment at their site. To capture tacit knowledge on machines as old as 50 years that lack proper documentation, the team records conversations among subject matter experts or retired employees and has a language model summarize them into documents that read like manufacturer manuals, which then become available to every application using the model. At the time of the case study the company planned to extend ChatGP to more of its facilities by the end of 2024.