Building and Managing High-Performing AI Teams in the Energy Industry: Lessons from the Embedded AI Lab

We created the Embedded AI Lab in 2018, using the term “Embedded” to reflect it’s position in the SLB Riboud Product Center (Clamart, France). The location was chosen to stimulate interactions between data scientists and engineers designing tools and software products. The team is now composed of 20 data scientists, who have expertise in three AI domains—computer vision, time series, and natural language processing (NLP)—and delivers products for all of SLB. After four years I would like to share some lessons learned during this journey, on the construction of the team, the organization, and the management of our projects.

Building a winning team
To succeed in the competitive world of AI and machine learning (ML), it's crucial to attract and retain top talent. Here are some strategies to consider when building your AI team:

  • Target the right profiles
    Focus on candidates with a strong appetite for learning, curiosity, and the ability to delve deep into new topics. Look for individuals who possess knowledge and aptitude in both AI and fundamental sciences, particularly applied mathematics.
     
  • Embrace diversity
    Create a dynamic and creative atmosphere by selecting team members with diverse backgrounds, including gender balance, nationality, and a mix of PhD and master's degrees. This diversity brings fresh perspectives and sparks innovation.
     
  • Highlight unique opportunities
    Attract candidates by showcasing the unique opportunities your organization provides. Highlight the rich scientific heritage and expertise of your company, as well as the potential for young data scientists to make a profound impact on the energy industry. Emphasize the range of projects, diverse data sources, and operational constraints they can expect to work on.
     
  • Foster professional development
    We implemented a de-risking phase called the "spike" before launching projects with a proof of concept and handover stage. During this phase, a group of two to three data scientists volunteered to initiate, brainstorm, and implement preliminary methods. This approach exposed team members to a wide range of projects, providing valuable opportunities for learning and growth. By actively participating in these spikes, our team members gained hands-on experience and expanded their knowledge base.

Lessons learned on project management
Throughout our journey, we've gathered invaluable lessons on project management. Here are a few key takeaways:

  • Collaborate with domain experts
    Mitigate risks in data science projects by working closely with stakeholders who possess domain expertise. Their involvement ensures proper data collection, understanding, and alignment with business objectives. It also facilitates knowledge transfer and fosters easier adoption of AI models.
     
  • Assess digital readiness
    To avoid wasted efforts, assess the digital readiness of a project before investing significant resources. Incorporate criteria to measure project maturity, ensuring alignment between AI and digital targets.
     
  • Enhance industrialization support
    To accelerate integration and support engineering teams expand your AI team's software expertise, especially in microservice development and MLOps. This enables smoother handover and industrialization of AI models and algorithms.
     
  • Embrace user-centric design
    Develop UI/UX skills within your team to implement quick mock-ups, providing domain experts with customized software to interact with. Early feedback and customization contribute to future software success and continuous model improvement. Simple interactive products can be implemented easily thanks to AI frameworks. [RJ1][OR2] The development could be done in a few hours!
     
  •  Stop a project when needed
    Throughout the life cycle of an AI project, it's important to acknowledge that challenges can arise—despite our best efforts! Periodically reassess and potentially pause or hold a project if needed. Taking this approach is a win-win situation as it prevents wasted time and safeguards collaboration and motivation within the team. The beauty of this approach is that it allows us to revive a project if we overcome the obstacles that were previously observed. By being adaptable and responsive to the project's needs, we ensure a more efficient and successful outcome in the long run.

The science behind success
Building impactful AI solutions requires adapting methodologies and combining traditional and modern approaches. Consider the following:

  • Prioritize high efficiency
    Simplify tasks and prioritize high model efficiency over handling complex or sophisticated tasks. An example is the augmentation of physics models or interpretation algorithms instead of replacing the existing ones, you can achieve impactful results more efficiently.
     
  • Combine traditional and modern methods
    Systematically employing deep learning is not always appropriate as the amount of data could be limited, or explainable AI could be required. Instead, spend time formalizing the problem, divide tasks into stages, and leverage appropriate methods for each stage. As an example, in one of our projects on image alignment, combining traditional image analysis techniques, deep learning, and dynamic programming yielded superior outcomes.

Looking ahead
In the energy industry, complex challenges demand innovation and pushing the boundaries of AI. Here are areas to explore:

  • Leveraging limited data
    Leverage physics and domain expertise to constrain models when data is limited. Also use mathematical knowledge and conventional methods. Lack of annotations or annotation consistency opens a wide field on unsupervised/self-supervised learning but also on noisy labels, and active learning.
     
  • Overcoming operational constraints
    Innovate to address operational constraints such as memory, computation capacity, telemetry, and connectivity. Innovate on methods to reduce model size for real-time inference (and probably training in the near future) while maintaining efficiency.
     
  • Collaborating with universities
    Strengthen and renew collaborations with universities to stay up to date with the latest advancements in AI. Engage in joint research to tackle challenging and exciting problems at the forefront of the data science community.

Conclusion
Building and managing successful AI teams in the energy industry requires strategic planning, collaboration, and a focus on innovation. By attracting the right talent, embracing diversity, implementing effective project management strategies, and combining traditional and modern approaches, you can position your organization for AI success. Stay committed to continuous learning, adaptability, and collaboration with academic institutions to tackle the evolving challenges in the field of AI for the energy industry.

Josselin Kherroubi
AI Director - Scientific Advisor

Josselin Kherroubi is a scientific advisor engineer, active in technical engineering and project management. He graduated from Centrale Supelec with a specialization in Applied Mathematics and a focus on image processing and machine learning. Since joining the SLB Riboud Product Center (SRPC) in Clamart in 2006, he has worked on automated fracture extraction and characterization, fracture stochastic modeling (Petrel reservoir modelling software), signal processing and inversion for dielectric interpretation software, and lead a team developing advanced image analysis modules (Techlog wellbore software platform).

He is currently leading the Artificial Intelligence (AI) laboratory located in SRPC, supervising a growing team, and accelerating new AI workflows for all the business divisions of SLB.

Please contact Josselin at Linkedin