CMS’s financial office is using LLM pilot to combat loss of institutional knowledge

For a federal agency charged with administering health insurance programs, upholding quality standards at facilities and enforcing a seemingly countless number of arcane rules, the Centers for Medicare & Medicaid Services is especially reliant on institutional knowledge. 

But with an aging workforce, occasional hiring freezes and an unyielding stream of work to manage, CMS’s financial leaders are turning to artificial intelligence to keep pace.

Joe Hong, director of CMS’s Division of Program and Data Management within its Financial Management Systems Group, said Wednesday during a Scoop News Group-produced UiPath public sector event that his office is running a large language model pilot that aims to “mitigate the loss of that institutional knowledge” while also “rapidly getting people up to speed.”

“With all of the potential generative AI and knowledge management, we’re working on a pilot … to basically give us precise, contextual-based answers,” Hong said during a panel discussion on empowering finance through AI and automation. “And it’s really … able to get information much sooner and stuff that’s credible and reliable.” 

In an interview with FedScoop after the panel, Hong noted that the Office of Financial Management is “not a public-facing program administering Medicare,” but it plays a critical behind-the-scenes role in helping others within CMS to process claims and ensure recipients receive proper benefits. 

“The best thing that we can do is equip the components that are more front-facing and getting them information, data reports that are much more accurate in a faster amount of time,” Hong said. “Therefore, they can make decisions quicker, and they can see what’s actually happening.”

Hong pointed to documentation on federally facilitated marketplaces as a “very technical” item that requires a bit of expertise to “translate” into “what does that mean for the accounting system.” Leveraging LLMs helps financial management staffers — especially those that are newer to the work — to “tie that together.” 

The pilot currently in use in Hong’s division is Meta’s Llama 2, described as “the next generation” of the tech giant’s open-source large language model. Hong said he’s preached an “LLM-agnostic” philosophy, noting that it doesn’t make sense to “be married to a specific LLM” when those systems are becoming more complex and capable by the day. But so far, Llama 2 has been successful thanks to CMS partners “providing solutions that give [the office] that flexibility.”

The Office of Financial Management workforce has quickly demonstrated a desire to “embrace” the technology, Hong said during the panel, and his CMS division “is inviting people to lead teams, bring the problems that they’re trying to solve, and give them the tools and the flexibility to work through them.”

Figuring out how best to utilize AI systems to support agency work will only become more crucial for federal financial staffers going forward, Hong added after the panel. 

“We don’t see AI or automation eliminating jobs,” Hong said. “It’s only going to create jobs because the work doesn’t go away, right? We just take on more complex problems.”

Written by Matt Bracken

Matt Bracken is the managing editor of FedScoop and CyberScoop, overseeing coverage of federal government technology policy and cybersecurity.

Before joining Scoop News Group in 2023, Matt was a senior editor at Morning Consult, leading data-driven coverage of tech, finance, health and energy. He previously worked in various editorial roles at The Baltimore Sun and the Arizona Daily Star.

You can reach him at [email protected].