Bevar Ukraine developed a virtual generative AI-powered assistant named Victor to aid Ukrainian refugees integrating into Danish society, utilizing AWS services for scalability and multilingual support.
The organization has been supporting Ukrainian refugees in Denmark since 2014, offering housing, job search, and integration services, as well as delivering humanitarian aid to Ukraine.
Challenges in refugee integration include accessing public services and language barriers, prompting Bevar Ukraine to create an AI system for continuous support and quality service.
AWS services like Amazon EC2, S3, and DynamoDB form the backbone of Victor's system, facilitating efficient conversation management, data storage, and natural language processing.
Anthropic’s Claude 3.5 and Amazon Titan Embeddings G1 enhance Victor's responses with advanced dialogue logic and semantic search capabilities, providing accurate and context-aware assistance.
Secure multilingual support is achieved through AWS services like Amazon Translate, AWS KMS, and Amazon Cognito, ensuring data privacy and secure customer identity management.
Building an AI assistant involves considerations such as model selection, response improvement, cost monitoring, embedding selection, and knowledge base optimization, all supported by Amazon Bedrock.
Technical enhancements include implementing machine learning for service coordination, human-in-the-loop validation, serverless architecture migration, and semantic search improvements.
The AI-powered system has significantly benefited Ukrainian refugees by saving time, improving support quality, and providing immediate assistance on public services and other inquiries.
This implementation serves as a model for creating impactful AI-powered humanitarian solutions, emphasizing the importance of cloud infrastructure and thoughtful system design for social impact.