Lumeris Elevates HR Operations with a Generative AI Chatbot Using Amazon Bedrock

Lumeris provides data-driven insights, technology, advanced risk management, and value-based care capabilities that empower health networks and physician groups to deliver high-quality care to patients, improve outcomes, and lower costs. In addition, Lumeris runs a Medicare Advantage health plan under the brand Essence Healthcare. Powered by the technology of Lumeris, Essence Healthcare continues to achieve high ratings for its commitment to service and quality from the Centers for Medicare and Medicaid Services.

In 2023, Lumeris began exploring the possibility of incorporating generative AI into its business. Its goal was to explore how this technology could help solve core business problems. At the same time, the company’s HR department—its People and Culture (P&C) team—was spending a lot of time answering employee questions, and Lumeris saw an opportunity to run this workflow more efficiently.

Lumeris built a generative AI chatbot, called Ask P&C, that answers employees’ HR questions. The chatbot is powered by a large language model (LLM) in Amazon Bedrock, a fully managed service that offers a choice of high-performing foundation models from leading AI companies. Lumeris began a proof of concept in January 2024, tested it with the P&C team, and rolled out Ask P&C to all employees in March 2024. “We took this from ideation to actual rollout in 3 months, which is amazing for new technology like this,” says Bajpai. “Using the out-of-the-box functionalities that AWS provides, we sped up rollout and implementation tremendously.”

Totaling more than 1,000 people, all the full-time employees at Lumeris can access Ask P&C to ask questions about benefits, time off, paid holidays, insurance plans, and other HR-related policies. Each week, the solution processes dozens of inquiries that would otherwise have been handled by a P&C staff member. This saves time for the P&C department, and employees can get responses immediately to most questions, instead of searching for the correct document or sending an email to the P&C team and waiting 1–2 days to get a response.

The chatbot has an accuracy rating of over 90 percent. It is built with a retrieval-augmented generation (RAG) architecture, which uses a knowledge base of P&C documents that is indexed by Amazon Kendra, an intelligent enterprise search service. (See Figure 1.) When someone asks Ask P&C, “What are our paid holidays?” the solution first pulls up relevant information about paid holidays from the P&C documentation and creates a context from the top results. Amazon Kendra can handle various file formats, and it includes a thesaurus feature that Lumeris can use to improve search results. For example, Lumeris can note that P&C means “people and culture” so that the solution will search for text containing either term. “For retrieval, Amazon Kendra was specifically an accelerator for us to be able to deliver a solution that works without a ton of effort on our part,” says Nick Dohr, director of engineering at Lumeris.

After retrieving context data using Amazon Kendra, the solution sends the user query and context data through the chosen LLM in Amazon Bedrock, which generates the response to the user. Lumeris uses Claude 3.5 Sonnet from Anthropic as the model for Ask P&C. “The key benefit of Amazon Bedrock was getting early access to powerful models to answer the complex inquiries that this solution requires,” says Dohr. “We tested and upgraded to Claude 3.5 Sonnet very quickly when it came out.”

With the architecture for the chatbot already in place, Lumeris can quickly implement additional use cases. In fact, Dohr’s team already provided the sales team with a proof of concept that took less than 3 business days to spin up. Using the RAG architecture, new chatbots can be trained on department-specific knowledge bases to answer employee questions or help quickly retrieve information from documentation.

By using AWS, Lumeris was able to launch this solution rapidly with minimal effort from its engineers. “Generative AI can be used to enhance human collaboration and team performance, and that’s exactly what we set out to do,” says Dohr. “Because AWS provides services that work really well out of the box, our teams were aligned to support this effort while still having the bandwidth to focus on the business priorities and operational needs at hand.”

Additionally, every Lumeris employee now has a chance to interact with generative AI and imagine where else it could be used to simplify their workflows. “A lot of employees have told me that until it was built, they did not realize how useful this tool could be,” says Bajpai. “With this tool, people can see the technology in action and interact with it, which gives a lot of credence to how the technology works and how they can use it in their daily lives and business functions going forward.”