Bevar Ukraine: Empowering Ukrainian refugees with AI-powered support on AWS

This is a guest post written by Bevar Ukraine, an AWS customer

When tens of thousands of Ukrainian refugees arrived in Denmark after Russia invaded Ukraine on February 24, 2022, they faced urgent questions about housing, healthcare, and legal rights—often while dealing with trauma and language barriers. In just three months, Bevar Ukraine developed an artificial intelligence (AI) virtual assistant that now helps volunteers support hundreds more people daily, creating a blueprint other humanitarian organizations can adopt.

Bevar Ukraine is an independent, Danish, non-profit humanitarian organization that delivers social and legal support to displaced Ukrainians as well as providing humanitarian aid to Ukraine. It has operated through volunteers since the 2014 launch who work closely with Danish public institutions, municipalities, commercial partners, and local communities.

AI helps manage thousands more queries

When demand for Bevar Ukraine’s services soared in 2022, volunteers repeatedly responded to similar issues while also managing complex, individual cases.

Traditional solutions such as static FAQ pages and basic chatbots proved insufficient. Refugees required conversational, context-aware assistance that could adapt to nuanced questions, was multilingual and could reflect any updates to Danish public services rules. At the same time, data protection, privacy, and trust were critical requirements.

The solution: Building Victor with AWS generative AI

To address these challenges, Bevar Ukraine collaborated with Amazon Web Services (AWS) to develop Victor, a generative AI-powered virtual assistant designed to scale access to reliable information while preserving a human-centered approach.

Agile development principles were crucial to success. The team launched a functional proof of concept and iterated based on real-world feedback. This approach allowed Victor to begin helping refugees quickly while continuously improving over time.

Human-centered design was a critical success factor. Throughout development, the team prioritized the refugee experience, ensuring that Victor’s interface was accessible, its responses were empathetic, and complex cases were escalated to human volunteers. Technology served as an enabler of human compassion rather than a replacement for it. A human-in-the-loop approach ensured continuous quality control, allowing volunteers to review responses, refine the knowledge base, and improve the system over time.

Similarly, data privacy and security considerations were integrated from the beginning rather than added as afterthoughts. Bevar Ukraine built GDPR compliance and data protection into the architecture so that refugees could trust the system with their questions and concerns.

Impact: Transforming refugee support at scale

Victor now supports hundreds of users daily, providing immediate answers around the clock on issues including housing applications, healthcare access, and employment registration.

In the first 150 days, the assistant has saved 1,500 volunteer hours by managing routine questions automatically, allowing the team to focus on complex cases that require human judgement, empathy, and follow-up. The team can support more people without proportionally increasing operational costs.

Methodology and tools

The project began with architectural workshops and close collaboration between Bevar Ukraine’s team and AWS specialists. AWS provided technical guidance, solution architecture support, and cloud credits to enable a proof of concept and rapid experimentation.

At the core of Victor is Amazon Bedrock, AWS’s fully managed service for accessing high-performance foundation models. The team selected Anthropic’s Claude 3.5 Sonnet on Amazon Bedrock for its strong contextual reasoning, conversational capabilities, and multilingual performance. Semantic search is powered by Amazon Titan Embeddings, enabling Victor to retrieve relevant information from curated knowledge sources, including Danish public authorities, social services documentation, and Bevar Ukraine’s own guidance materials.

The solution architecture combines  Amazon Elastic Compute Cloud (Amazon EC2) for compute, Amazon Simple Storage Service (Amazon S3) for secure storage, and Amazon DynamoDB for managing conversation state and interactions. Additional components, such as vector search and prompt optimization frameworks, were integrated to improve response quality and consistency.

From the outset, privacy and security were built into the design. Amazon Bedrock’s built-in security controls supported encryption, isolation, and responsible use of AI models.

Victor moved from concept to production in three months. AWS’s managed services reduced the need for infrastructure maintenance, allowing Bevar Ukraine to focus on content quality and user experience rather than operations.

Future plans include more advanced context routing, improved escalation logic, and further optimization of scalability and cost efficiency. The initiative led to the volunteer team establishing a generative AI start-up, CoreAI.

Lessons learned: A blueprint for humanitarian AI

The Victor assistant offers valuable lessons for other organizations seeking to leverage AI for humanitarian purposes. Digital preparedness proved essential—Bevar Ukraine’s willingness to embrace cloud technology and AI enabled rapid response to refugee needs. Cross-sector collaboration between a volunteer organization and a technology provider like AWS demonstrated how partnerships can accelerate innovation in the social sector. Organizations interested in implementing similar solutions can access a range of resources, such as the AWS Imagine Grant Program, and find more information at the AWS for Nonprofits webpage.