AI Connect: YMCA

The Arizona State University Artificial Intelligence Cloud Innovation Center, powered by Amazon Web Services (AWS), collaborated with YMCA of Snohomish County to design and deploy an AI-enabled multilingual chatbot that transforms access to YMCA historical knowledge. Using a Retrieval-Augmented Generation (RAG) pipeline on Amazon Bedrock and a fully serverless AWS architecture, the solution automates document digitization (OCR), semantic indexing, and real-time, citation-backed conversational responses—available in 12 languages with seamless bidirectional translation.

This solution helps YMCA staff, members, and community partners quickly discover relevant institutional knowledge from historical archives that were previously difficult to search due to unstructured formats and language barriers.

YMCA organizations maintain extensive archives—meeting minutes, program documentation, and community impact reports—containing valuable institutional knowledge. However, these records are often stored as physical documents or unstructured digital files, making discovery slow and resource-intensive. Language barriers further limit access for diverse communities, and manual search or multilingual support does not scale.

YMCA needed a solution to:

• Digitize historical documents reliably (including scanned PDFs, forms, and tables)
• Make archives searchable via natural-language questions
• Provide fast, multilingual access to information with source citations
• Reduce staff burden while improving accessibility and community engagement

The CIC team built an end-to-end knowledge management workflow on AWS serverless infrastructure:

• Document ingestion: Admins upload to S3 via Cognito. Step Functions orchestrates Textract OCR, supporting scanned PDFs up to 400 pages.
• Semantic indexing: Content is ingested into Bedrock Knowledge Base using Titan Text Embeddings V2 and S3 Vectors for vector storage and similarity search.
• RAG synthesis: Bedrock retrieval uses diversity controls to encourage multi-document grounding and avoid over-reliance on a single source.
• Multilingual support: Amazon Translate enables 12-language bidirectional translation with auto-detection.
• Response streaming: Token-by-token streaming via Lambda Function URLs and SSE reduces latency for complex queries.
• Analytics: Conversation history and metrics stored in DynamoDB, surfaced through an admin dashboard.

The YMCA AI Multilingual Chatbot modernizes knowledge access by converting static historical archives into an interactive resource that serves both internal staff and multilingual communities. This project demonstrates the potential of AI to modernize how organizations manage and share information. By digitizing unstructured archives, the chatbot lets users search for information using natural language, breaking down barriers that previously made these archives difficult to navigate. The inclusion of multilingual support ensures that diverse communities can engage with the YMCA's history, ensuring inclusivity and accessibility. The chatbot's design also reduces the workload for YMCA staff by automating tasks like document digitization and translation. This allows staff to focus on more impactful work, while the community benefits from faster and more accurate information. The solution's scalability and modular design means it can be adapted for other organizations facing similar challenges, such as nonprofits, educational institutions, and public-sectors.

We at the Y are truly grateful for our partnership with the Arizona State University Artificial Intelligence Cloud Innovation Center. This project and collaboration is helping make 175 years of YMCA institutional knowledge and history more accessible to Y staff, volunteers, and the communities we serve. Working with the talented student leaders and staff at the ASU CIC has been both informative and inspiring. Collaborations like this are essential for responsible deployment of AI technologies in our communities.

Michael Rinner - Sr. Dir. of Technology, YMCA of Snohomish County

This architecture can be extended beyond YMCA historical archives to support broader knowledge management use cases, including:

• Program documentation and policy knowledge bases
• Multilingual community Q&A portals
• Internal staff enablement across distributed teams
• Other nonprofit and public-sector organizations with unstructured archives

The solution is designed to scale through serverless components, modular workflows, and 

configurable retrieval and translation settings.