Modernization AI Assistant: NASWA

The Arizona State University (ASU) Artificial Intelligence Cloud Innovation Center (AI CIC), powered by Amazon Web Services (AWS), collaborated with the Unemployment Information IT Support Center (UI ITSC) offered by the National Association of State Workforce Agencies (NASWA), in partnership with the U.S. Department of Labor (USDOL)— to prototype a multilingual, generative AI chat assistant for UI IT modernization. The solution transforms how UI ITSC staff and state workforce agency professionals can access the organization's extensive technical documentation, enabling faster, more accurate guidance across all 50 states, D.C., and the U.S. territories.

UI ITSC in partnership with USDOL serves as a national resource for state unemployment agencies, providing IT guidance and facilitating knowledge sharing as states modernize their IT systems. Central to this effort is the UI IT Modernization Guidebook, a roughly 1,000-page reference document containing deep technical and programmatic content across years of accumulated expertise. However, the sheer size of the Guidebook makes it difficult to navigate. Users often must manually search through extensive documentation to find relevant information, a process that can be both time-consuming and inefficient.

The CIC team designed and built a Retrieval Augmented Generation (RAG)-based chatbot prototype on AWS that supports multilingual (English/Spanish) Q&A over UI ITSC’s content, with a web user interface delivered as a pop-up chat experience.

• AI/RAG: Amazon Bedrock Knowledge Base retrieves relevant context from ingested materials and supports configurable model selection for response generation.
• Embeddings & Retrieval: Amazon Titan / Nova embeddings transform content into vectors, enabling semantic search over the knowledge base.
• Knowledge ingestion: Amazon S3 stores source content and processed artifacts; ingestion supports HTML, Excel, Word, and PDF formats, with Amazon Bedrock Data Automation used for document parsing/structuring.
• Web & content collection: A web crawler gathers relevant site content for indexing and refresh, feeding the pipeline for embedding and storage.
• Backend orchestration: Amazon API Gateway fronts REST/HTTP endpoints to connect the web UI to backend request handling (e.g., an “agent handler” service) that routes queries to the knowledge base and selected Large Language Model (LLM).
• Frontend experience: A browser-based UI (hosted via AWS Amplify) provides a simple chat interface intended for desktop use and supports multilingual interactions.
• Access control: Amazon Cognito can be used to authenticate users prior to accessing the chatbot experience.
• Conversation context: An agent memory layer can persist chat history (e.g., in S3) to enable more consistent multi-turn experiences.
• Document Source Attribution: Every response includes relevance scoring and direct links to source documents, supporting transparency and user trust

This prototype demonstrates how NASWA and state workforce agencies can improve access to UI information by providing a consistent, intelligent self-service channel grounded in approved program documentation. The virtual assistant directly solves the Guidebook navigation problem-- what once required manual searching through hundreds of pages can now be answered in seconds through a natural language query. For state IT staff and program administrators, this means faster access to the guidance needed to make informed modernization decisions. 

Equally important, the solution serves as a reference implementation that states can learn from as they build their own AI capabilities. By sharing this prototype with member states, UI ITSC can reduce duplicated effort, accelerate AI adoption, and help states make more informed decisions when designing their own AI solutions.

“Partnering with AWS and the ASU AI Cloud Innovation Center enabled us to develop a working prototype that addresses a real operational challenge—making decades’ worth of technical guidance accessible through a simple, conversational interface. Beyond the immediate use case, this project has provided a clearer understanding of how AI can support knowledge transfer across our organization and the states we serve. It has also meaningfully influenced how we approach AI adoption moving forward.”

Brett Flachsbarth | Director, Unemployment Insurance IT Support Center | National Association of State Workforce Agencies

Because the solution is designed as a reusable framework with customizable LLMs and knowledge bases, it can be adapted across a wide range of internal and external use cases. The same architecture can support future knowledge bases and be tailored to the needs of individual state agencies or other divisions within NASWA. Potential applications can include RFP interrogation and creation, Employment and Training Administration (ETA) report processing, and Benefit Accuracy Measurement (BAM) staff training. As UI ITSC's communities of practice and state partners contribute additional content and use cases, the platform has the potential to grow into a shared national resource for UI innovation.