How We Collaborated with Baker Hughes & AWS to Create Novel Digital Assistants & Build a Repeatable Framework for GenAI Initiatives
Leveraging GenAI to Increase Productivity
At a glance
CLIENT
Baker Hughes
STRATEGIC PARTNER
- AWS
SERVICE
- Artificial Intelligence
INDUSTRY
- Energy & Resources
In collaboration with Baker Hughes and Amazon Web Services (AWS), EPAM quickly built and piloted two digital assistants to drive increased productivity and provide enhanced insights to users. The successful delivery of these projects allowed us to create a repeatable framework that can quickly and effectively deliver future generative AI (GenAI) initiatives.
Baker Hughes is a leading energy technology business and one of the largest oilfield services companies in the world.
Recognizing the power of artificial intelligence (AI) to deliver new value, Baker Hughes turned to EPAM, a longtime collaborator, to support the development of two digital assistants, one for production engineers and another for field engineers. Both digital assistants were developed using the robust resources of the AWS GenAI toolkit.
Helping humans do what they do better is the foundation for both GenAI collaborations between Baker Hughes and EPAM. These projects successfully demonstrate how to enable GenAI in a large energy company, which means Baker Hughes not only has two new digital assistants to use, but it also has a framework for how to build and execute additional GenAI projects going forward.
DIGITAL ASSISTANT FOR PRODUCTION ENGINEERS
Making Data More Accessible and Useful
Built in collaboration with EPAM and AWS, Leucipa™ is an automated field production software solution using the power of AI prediction models. With the help of machine learning, Leucipa analyzes vast troves of data to predict maintenance needs for wells so that downtime is minimal or avoided altogether. It’s a gamechanger for Baker Hughes’ clients.
Baker Hughes wanted to leverage generative AI (GenAI) by building a conversational digital assistant to complement Leucipa. Using the rich set of production and operational data available via Leucipa, this intelligent assistant would uncover new investigation patterns for production engineers, cutting the time it takes them to reach critical insights.
In just four weeks, a small team of EPAM engineers, designers and energy consultants delivered a pilot GenAI digital assistant to help production engineers with questions related to oil production, well performance, well treatments, production targets and artificial lifts.
In a matter of seconds, the Leucipa digital assistant can report current production levels, pinpoint wells producing below target, return the date of the last artificial lift treatment, advise on the average volume of acid used in a particular field and more. It can also generate SQL queries corresponding to the information being requested and create interactive plots for a better analytics experience.
All of this was built using AWS cloud capabilities with extensive use of Amazon Bedrock in particular. Amazon Bedrock allowed the team to iterate through several foundational GenAI models and select the best candidate (Anthropic’s Claude 3.5) for optimal performance. The flexibility of the AWS toolkit made the Kendo and HighCharts-driven UI components perform at their best.
Baker Hughes is now introducing the Leucipa digital assistant to its clients with plans to develop a production-ready release of the GenAI assistant in the near future.
DIGITAL ASSISTANT FOR FIELD ENGINEERS
Driving Internal Efficiencies
The Artificial Lift Systems (ALS) Team, a group of experts within Baker Hughes, answers technical questions from field engineers working with Baker Hughes equipment across the globe. The ALS Team receives 1,600 requests for support annually, with the number growing.
The pain point to be addressed with a new GenAI-powered digital assistant was straightforward: Significantly reduce the estimated 10,000 hours that experts spend annually answering routine, repetitive questions about equipment. With field engineers using GenAI for simple questions, human experts would be freed up to tackle more complex challenges.
To build the ALS digital assistant, we processed a range of documents, including technical bulletins, handbooks, catalogs and operational procedures using Amazon Textract. We created embeddings using Amazon Titan models from the Amazon Bedrock tookit and stored them in Amazon OpenSearch.
Using both keyword and similarity search approaches with Anthropic’s Claude models, available via Amazon Bedrock, and after multiple rounds of testing and refining with subject matter experts (SMEs), we delivered a pilot digital assistant that could answer a broad spectrum of questions with a remarkable 85% accuracy. The extensive toolkit offered by AWS for GenAI capabilities allowed us to iterate, try out different methods and deliver a working pilot in record time.
The next step is for the prototype to be scaled to include more documents, so it can answer a wider range of more complex questions. Then it will be integrated into the ALS portal, so it can be accessed directly by field techs globally.
Results
8
Weeks to deploy a fully functional prototype
200+
Pages of technical documents processed
85%
Success rate on answers
CHARTING THE FUTURE
Executing a Repeatable Framework for GenAI Enablement
In developing these two digital assistants, Baker Hughes now has expertise in implementing real-world GenAI tools. These experiences highlight a repeatable framework for developing a wide array of GenAI tools that can provide efficiencies and cost savings across business units well into the future.
A GenAI Enablement How-To Guide:
Finding the Right Use Case
Finding the Right Use Case
Value doesn’t have to be flashy. Finding the right problem to solve means understanding what data can do and where it can add incremental value. As we evaluate multiple use cases, a key question is around what data is both accessible and useful. The value of any GenAI initiative is defined by the quality of data supporting it and by the savings derived or revenue generated by it.
Assembling the Right Team
Assembling the Right Team
It’s critical to have the right cross-functional team to successfully deliver GenAI projects. A group of skilled developers is necessary but insufficient by itself — the team must include GenAI technologists (engineers, developers, architects), domain SMEs and business-process consultants to ensure successful execution. It’s also essential to have representation from the business as project sponsors. That’s the only way to reliably deliver on value targets.
The Baker Hughes SMEs were vital to validating outputs and helping to drive adoption. With the ALS digital assistant, the project sponsor had a vision of how GenAI could drive value and promoted the digital assistant, driving engagement.
Understanding Key Requirements
Understanding Key Requirements
Each organization has its own restrictions and requirements around security and data. It’s important to understand that environment up front to facilitate the correct technical design. Specifically, stakeholders must align on expectations regarding data privacy, hallucinations, validation of results and approved technologies. This "pre-work" directly contributes to the design and overall success of the project, plus it helps avoid delays or costly rework.
Readying Your Services
Readying Your Services
There is prep work that should be done before launching an AI project to ensure a smooth transition and execution. Baker Hughes now has experience in checking off those steps: ensuring the right technology is approved (e.g., LLMs, knowledge graphs), gaining necessary approvals from internal GenAI governance, onboarding developers to be ready on Day 1 and ensuring the environment is provisioned to run pilots.
Piloting, Iterating, Improving
Piloting, Iterating, Improving
It’s critical to incorporate real user feedback to refine any tool. Trial and error, with prompt feedback from SMEs, allowed for rapid continuous improvement during development. And because the SMEs were seeing progress, they were more confident that the effort was worthwhile, which inspired them to keep development moving.
Putting GenAI to Work
There’s a lot of hype around GenAI, but the bottom line is it can be an invaluable tool when designed and implemented wisely. Understanding how to get the most out of your data, putting the right team in place, utilizing the best tech for the project and employing continuous improvement are the hallmarks of a smart GenAI initiative.
HEAR FROM THE CUSTOMER
“Through our collaboration with EPAM, we were able to assemble hybrid teams and create new complex models for advanced predictive failure for electric submersible pumps. And now, thanks to our work together, we’re generating millions of dollars of customer value across conventional and unconventional wells. Having the ability to adapt to different methodologies and being truly agile guided us to a real success in this project.”
Sebastiano Barbarino
Leucipa Director of Product & Engineering, Baker Hughes
LEARN MORE
Partner With Us
Explore our expertise in the energy sector.