DSI Consulting Service spurs innovation
Today, research in nearly every scientific discipline involves data science techniques. Whether using sophisticated tools to manage and analyze massive datasets or applying machine learning algorithms to gain new insights, such techniques are becoming ever more prevalent. However, scientists and engineers may not have specific training in the newest, most pertinent data science and statistical methods that could boost their projects.
Recognizing this gap, the Data Science Institute (DSI) launched the DSI Consulting Service, DSICS (pronounced “D6”), which temporarily places Livermore data science consultants with internal projects that would benefit most from their expertise. Consultants work directly with a research team and advise on statistical methods, data science tools, and data management approaches best suited to a project’s needs.
“DSICS is perfect for filling small, short-term gaps in research projects. The Laboratory’s demand for this type of service is large, and it’s growing. Many early- to mid-career consultants have signed up from across organizations including Physical and Life Sciences [PLS], Engineering, Computing, and beyond,” says DSICS director and statistician Jason Bernstein. Last year, the program brought on three deputy directors—Gemma Anderson, Giselle Fernandez, and Aldair Gongora—to assist with scaling up. Bernstein believes that with more consultants and more project requests, the DSICS program’s impact at the Laboratory will continue to expand.
Building a Consulting Foundation
“More and more of us who don’t necessarily work exclusively in data or computer science fields are gaining data science knowledge; we’re technicians, business analysts, or administrators,” statesTyler Alcorn, a health physicist in LLNL’s Radiation Protection Functional Area and DSICS consultant. Alcorn, who has a background in nuclear engineering technology, came to Livermore as he was just beginning a master’s degree in applied data science.
As he progressed in his degree program, Alcorn appreciated how the data science techniques he was picking up through academics and personal projects could play a role in his radiation protection work, too. Health physicists regularly deal with large volumes of environmental testing data which they analyze to determine a work area’s radiological hazards and to implement appropriate safety protocols. They also consult historical records that detail the previous types of work and experimentation performed in a location. Piecing together these disparate data sources with historical information is tedious and time-consuming, but Alcorn saw the opportunity to try out new methods.
“The health physics profession was born out of the nuclear power plant industry. Its operations are more old-fashioned to align with strict regulations, so it has been slower than other disciplines to incorporate certain technologies. For instance, lots of information is handwritten,” Alcorn explains. “One of the ways I started applying data science techniques at Livermore was by creating a vector database of historical environmental testing and usage information to enable the tracking, trending, and analysis of health physics data. I also used Monte Carlo simulations to predict radioactivity doses under different environmental conditions. Now, instead of having to rely on legacy institutional knowledge, I could, for example, determine that a particular area was associated with National Ignition Facility research and dealt with tritium—which can be dangerous if ingested—and better understand the risks going forward.”
While working on health physics applications, Alcorn attended a DSICS road show held to raise awareness of the program and spur consultant sign-ups. After joining the cadre of consultants, he was contacted for a project that expanded on his knowledge of building vector databases, the foundation of large language models (LLMs). The engagement was so successful that Alcorn now spends half of his time working alongside DSI deputy director Cindy Gonzales on retrieval augmented generation (RAG), using LLMs to parse historical documents and quickly provide requested information and insights. “Being a DSICS consultant was a perfect way for me to get my feet wet in new topics and practice by doing. I absolutely recommend taking part, either by using the service or becoming a consultant yourself,” says Alcorn.
Award-Winning Ideas
Josh Ottaway, the DSI’s ambassador to LLNL’s Strategic Deterrence (SD) organization, has experience on the opposite side of the customer-consultant relationship. Though his home base is PLS, Ottaway works primarily with SD to ensure the safety, security, and reliability of the U.S. nuclear arsenal as part of the stockpile surveillance program.
“We have lots of opportunities in SD to use data science more than we do currently. Machine learning and artificial intelligence in particular are promising assets,” he says. According to Ottaway, the Laboratory has leaned into efforts to perform model-based assessment of components through constructing digital twins. And, in a similar vein to Alcorn’s work, the organization could benefit from specialized LLMs to sort through extensive classified records, retrieving and interpreting information more efficiently than by hand.
In early 2024, Ottaway brought on two DSICS consultants, Katie Schmidt and Laura Wendelberger, to assist with an SD effort related to reliability reporting statistics. Reporting on stockpile reliability is a significant deliverable for Livermore accomplished by assessing individual components, reviewing experimental data, and performing predictive simulations.
Schmidt explains that they “introduced the concept of Bayesian reliability and produced R scripts for estimating reliability and quantifying the associated uncertainty.” The pair also helped develop reliability content for a study by JASON, a board of independent scientific experts that reviews specific topic areas for the nuclear security enterprise. For their work, each was recognized with a Silver Award from SD’s Weapon Technologies and Engineering Program.
DSICS consultants have also earned recognition from other areas of the Lab. Postdoctoral researcher Mike Boyle was recognized with a Bronze Award from Global Security for his work supporting the Forensic Science Center (FSC). FSC sought to calculate the masses of millions of different molecular compounds with different atomic structures. Boyle devised a Python script that iterated through combinations of functional groups present in each molecule, saving hundreds of hours of labor in generating a comprehensive screening library for the international Organisation for the Prohibition of Chemical Weapons.
Bernstein is excited that participation in the DSICS program continues to grow along with its successes. “We’ve built a sizable base of consultants, and they’re really eager to help. Project teams can come to us at any point, even before they’ve begun collecting data. We want to ensure they have the right kind of data to test hypotheses and answer questions,” he notes. For consultants, branching out also provides experiential benefits. Bernstein adds, “DSICS is a great avenue to network and offer technical impact beyond your usual role, and ultimately, it helps us achieve our mission.”
