AWS and Bevar Ukraine build a multilingual support system for displaced people
In the wake of humanitarian crises, timely and accurate information can be as vital as food or shelter. But for refugees navigating an unfamiliar country, that information is often locked behind language barriers, bureaucratic complexity, or simply a lack of available personnel. Now, a collaborative effort between Danish NGO Bevar Ukraine and Amazon Web Services (AWS) is showing how generative AI can offer multilingual support at scale—and with empathy.
The project centers around Victor, a virtual assistant designed to help Ukrainian refugees in Denmark access crucial services: from housing and legal advice to education and healthcare. What makes Victor stand out is not just its scalability, but its localization-driven architecture.
A Translation Layer Beyond Language
Built using Amazon Bedrock, the assistant uses Anthropic’s Claude 3.5 model for conversational fluency and Titan G1 embeddings to contextualize user queries. This combination enables Victor to understand multiple languages and retrieve relevant content from a knowledge base tailored to refugee needs.
AWS services like Amazon Translate, API Gateway, and DynamoDB handle real-time interactions, while compliance tools ensure that privacy and GDPR standards are met. The result is a scalable backend that supports multilingual, 24/7 conversations without over-relying on human volunteers.
But this isn’t just about speech recognition or auto-responders. What AWS and Bevar Ukraine have built resembles a localized AI agent—one that can switch languages, respect cultural nuance, and maintain functional parity across contexts. It’s the kind of automation the localization industry has long imagined but rarely seen fully realized.
Multilingual AI with a Mission
Victor is already handling hundreds of interactions per day, offering support in multiple languages. Because it’s powered by vector embeddings and trained on domain-specific content, its responses stay contextually accurate, even across different linguistic inputs.
More importantly, it’s a blueprint for how AI-powered localization can become a pillar of humanitarian tech. For developers, translators, and linguists alike, Victor represents a convergence point where ethical AI, machine translation, and user-centric design meet to serve real human needs.
As language access becomes a right, not a privilege, projects like this one show how the infrastructure of empathy can—and must—be multilingual by design.
