Multilingual Resource Finder: USCRI

The Arizona State University Artificial Intelligence Cloud Innovation Center, powered by Amazon Web Services (AWS), collaborated with the U.S. Committee for Refugees and Immigrants (USCRI) to explore a prototype multilingual, AI-powered chatbot that provides accessible, location-based community resource guidance, examining how conversational interfaces could improve access to trusted information for immigrant and refugee communities with diverse language needs and varying levels of digital literacy.

Built as a proof of concept, the solution delivers real-time assistance in 11 languages, automatically searching listings from USCRI’s community resource directory. By combining language detection, translation, conversational intelligence, and location-aware resource recommendations, the chatbot enables immigrant families, unaccompanied children, sponsors, and service providers to navigate critical services with greater clarity and confidence—while maintaining a privacy-first experience with no conversation or PII retention.

One of the many ways USCRI serves immigrant and refugee communities nationwide is by providing information to individuals, families, and service providers on free and low-cost local resources through its Community Resource Directory, which includes more than 45,000 verified listings across the United States. As USCRI continues to expand access to these resources, it recognizes that users approach digital tools with different preferences, comfort levels, and access needs.

Some users, particularly those navigating high stress situations, language barriers, or unfamiliar digital environments, may benefit from guided, conversational ways of accessing information alongside traditional directory search and browsing tools. This project focused on exploring whether a multilingual chatbot interface could serve as an additional access modality to complement USCRI’s existing user experience.

The CIC team built a privacy-first, mobile-friendly multilingual resource-discovery prototype chatbot for USCRI’s Community Resource Directory, with three main flows: Visitor Chat, Resource Export, and Directory/Data Integration.

• AI + language: Amazon Bedrock for response generation and intent/slot filling; Amazon Translate for automatic language detection + translation so users can interact naturally across commonly spoken immigrant languages (starting with English/Spanish and expandable).

• Directory lookup (real-time ready): A serverless “Directory Connector” API (dummy implementation in the POC) that simulates calling USCRI’s internal directory service to return curated listings (address, phone, hours, services) for the user’s requested category and location.

• Location-aware recommendations: Browser-permitted geolocation (when allowed) or manual ZIP/city/state entry is used to rank and filter nearby resources; the chat flow asks lightweight clarifying questions when the request is ambiguous (e.g., “Healthcare” vs. “I’m not sure”).

• PDF export (multilingual): A serverless export service generates a downloadable PDF resource list in the user’s selected language using USCRI’s templates, enabling users to save/share information offline.

• Privacy + temporary storage: Resources selected for export are stored temporarily (24-hour TTL) only to support PDF generation; no conversation history or location history is retained to align with privacy requirements.

• Frontend: A responsive web UI designed for mobile access and accessibility, embedded alongside the Community Resource Directory experience to reduce digital literacy barriers and speed up discovery.

The Multilingual Resource Chatbot prototype highlights how conversation and multilingual AI technologies can complement large, trusted resource directories by offering users a different way to engage and search listings.  The solution supports USCRI by breaking down language barriers across 11 languages, delivering personalized location-based service recommendations, reducing friction for users with limited digital literacy, enabling faster access to free or low-cost community services, and maintaining user trust through a privacy-first, no-PII storage model. By simplifying how immigrants and newcomers discover and access support services, the chatbot could contribute to more equitable service delivery and a more welcoming, accessible system of care.

“Working with the ASU Cloud Innovation Center gave USCRI the opportunity to explore how emerging AI technologies might support multilingual access to community resources. The prototype helped us think more intentionally about conversational interfaces and accessibility as part of our broader digital strategy.” 

Gregory Bailey, Program Manager, UC Resource Center, USCRI

While initially designed for resource navigation for USCRI’s Community Resource Directory, this solution architecture can scale across additional public-service contexts, including municipal and state human services portals, nonprofit resource directories, community legal aid navigation tools, and multilingual crisis and referral support systems. The modular serverless design enables future expansion into production deployments, deeper API integration, enhanced security controls, and broader language coverage.