When job loss strikes, the ability to quickly and easily access unemployment benefits can mean the difference between financial security and hardship. Yet for many government agencies, delivering fast, high-quality customer services at scale remains a challenge. In North Carolina, the Department of Commerce’s Division of Employment Security (NCDES) is tackling that challenge head-on with the help of Amazon Web Services (AWS).
Building on an eight-year cloud transformation journey with AWS, NCDES has launched one of the first generative artificial intelligence (AI)-powered, public-facing solutions in the state: an intelligent virtual assistant that helps individuals navigate the unemployment insurance claim application process.
Developed in collaboration with AWS and AWS Partner Ernst & Young, the solution demonstrates how thoughtful generative AI adoption can enhance citizen services while improving operational efficiency.
Supporting North Carolinians with transformative cloud modernization
The NCDES administers North Carolina’s unemployment insurance system, delivering essential services to individuals and employers across the state. Historically, the agency relied on traditional systems and manual processes to manage claims. Recognizing the limitations of aging infrastructure, NCDES embarked on a cloud-first strategy with AWS in 2018.
“Eight years ago, our footprint in the cloud was zero,” said Raju Gadiraju, chief information officer (CIO) at NCDES. “Today, more than 80 percent of our workloads are deployed in the cloud. That’s a big change, and it has put us in a position to take advantage of new technologies to better serve our customers.”
Modernization proved critical during the COVID-19 pandemic, when NCDES deployed remote desktop solutions, was the first agency in the state to adopt a cloud-based call center, launched eight new unemployment insurance programs within two months, and distributed $14 billion in benefits—all powered by cloud scalability. This transformation also paved the way for the agency’s next step: integrating generative AI to further enhance its citizen services.
Identifying the right use case for generative AI
For NCDES, the potential of generative AI to support citizen services was clear—but so were the risks. “Everybody is really scared of generative AI, especially if it’s public-facing,” said Gadiraju. To move forward responsibly, NCDES prioritized a pilot use case that solved a real business need to improve the customer’s experience, but was not at risk of exposing personally identifiable information (PII) and did not involve decision-making.
The division identified customer service—specifically the initial claim filing process—as a natural starting point for improvement. Traditionally, individuals with questions relied on the support center for assistance, facing wait times during peak periods and limited access outside of regular business hours. By deploying a generative AI-powered chatbot grounded in the agency’s centralized knowledge base, NCDES could provide consistent answers 24/7—improving accessibility while reducing call volumes.
“We asked ourselves, why are people calling? How can we improve the experience, so they don’t have to?” remarked Gadiraju. “We wanted the chatbot experience to be as good as talking to a person and available in real time.
Building a secure and scalable generative AI chatbot solution on AWS
Because NCDES’s core unemployment benefits system already operates in AWS GovCloud (US), an isolated US sovereign partition that supports highly regulated organizations like federal and state agencies, building the chatbot solution on AWS offered a natural extension. Familiarity with AWS services, combined with the flexibility of Amazon Bedrock to experiment with multiple AI foundation models, made AWS the preferred choice.
Working closely with AWS and AWS Partner Ernst & Young, NCDES launched the solution in just over four months. The architecture includes:
Security and responsible AI use were central throughout the project. NCDES mapped its implementation to North Carolina’s Principles for the Responsible Use of AI and the National Institute of Standards and Technology (NIST) AI Risk Management Framework. By grounding chatbot responses in a curated internal knowledge base, applying guardrails to limit hallucinations and maintain strict adherence to domain subject matter, and masking any inadvertent PII, the agency created a solution that protects both customers’ privacy and public trust.
“Ultimately, we didn’t treat this as a technology project,” commented Gadiraju. “It’s a customer experience project tied to a business need.”
Solution overview
Figure 1. Architecture diagram of the generative AI chatbot solution
This generative AI-powered chatbot architecture—shown in the preceding figure—combines traditional FAQ matching with advanced AI capabilities in a seamless flow, delivering accurate responses through the following process:
The user sends input to the SCUBI chatbot, which forwards it to Amazon Lex for natural language understanding and intent processing.
Amazon Lex sends a fulfillment request to AWS Lambda, which first queries Amazon OpenSearch Service for matching predefined FAQs.
If OpenSearch finds relevant FAQ matches, they’re returned through Lambda and Amazon Lex to the user. If no matches are found, Lambda activates the Amazon Bedrock fallback process.
When using Amazon Bedrock, the system queries its knowledge bases, which retrieve context from vector embeddings stored in OpenSearch (originally created from S3 data sources).
The foundation model in Amazon Bedrock processes the user prompt with this retrieved context and sends its response back through Lambda and Amazon Lex to the chatbot, which delivers the final answer to the user.
Throughout this process, user interaction data is collected to monitor trends and optimize the system through the analytics dashboard.
Early insights and lessons learned from launching generative AI
The NCDES generative AI chatbot went live in February 2025, and while it is still early, the agency is already seeing positive indicators. The chatbot handled 2,700 inquiries in its first month that would otherwise have required human intervention. Internal dashboards continue to track usage and key topics, while also providing rich insights into user behavior and potential areas for process improvement.
“I’m fascinated by how much insight we have,” said Gadiraju. “We can see what kinds of questions are being posed and if there’s a pattern across these questions. For example, are these questions highlighting inadequacies in the system itself? If we redesigned a certain screen, maybe people wouldn’t have that question anymore.” NCDES is looking forward to using the chatbot question data to improve both its knowledge base and its online claims process.
The chatbot also enhances the employee experience. Business users can review chatbot interactions, trace responses back to specific source documents, and update content as needed—creating a continuous improvement loop that benefits both individuals and staff. Plus, as the chatbot reduces call volumes off of support center staff, these employees can pivot their resources to support higher skilled tasks.
Paving the way for future generative AI innovation
For agencies considering their own generative AI initiatives, Gadiraju and Crystal Pitts, chief information security officer (CISO) at NCDES, recommend starting small with a clearly defined, low-risk use case. “Choose a project that doesn’t involve PII or decision-making,” said Pitts. “And focus on building a secure, responsible foundation you can build on over time.”
For NCDES, this generative AI chatbot is only the beginning. The agency is already exploring opportunities to extend AI-driven support to other areas, such as enhancing public-facing resources and assisting support center staff with knowledge management.
“It’s not only about customer experience; it’s also an employee experience improvement,” said Pitts. “We’re excited about efficiency and process improvement—and ultimately, better service for North Carolinians.”
AWS supports government agencies through every phase of their generative AI journey. Contact us today to learn more.