Leveraging AI for Social Impact: Building the Victor Assistant for Ukrainian Refugees
Introduction
This post is co-written with Taras Tsarenko, Vitalii Bozadzhy, and Vladyslav Horbatenko. As organizations worldwide seek to use AI for social impact, the Danish humanitarian organization Bevar Ukraine has developed a comprehensive virtual generative AI-powered assistant called Victor, aimed at addressing the pressing needs of Ukrainian refugees integrating into Danish society. This post details our technical implementation using AWS services to create a scalable, multilingual AI assistant system that provides automated assistance while maintaining data security and GDPR compliance.
Background and Challenges
The integration of refugees into host countries presents multiple challenges, particularly in accessing public services and navigating complex legal procedures. Traditional support systems, relying heavily on human social workers, often face scalability limitations and language barriers. Bevar Ukraine’s solution addresses these challenges through an AI-powered system that operates continuously while maintaining high standards of service quality.
Solution Overview
The solution’s backbone comprises several AWS services to deliver a reliable, secure, and efficient generative AI-powered digital assistant for Ukrainian refugees. The team consisting of three volunteer software developers developed the solution within weeks.
Integration and Enhancement Layer
Bevar Ukraine has extended the core AWS infrastructure with several complementary technologies:
- Pinecone vector database – For efficient storage and retrieval of semantic embeddings
- DSPy framework – For structured prompt engineering and optimization of Anthropic’s Claude 3.5 Sonnet responses
- EasyWeek – For appointment scheduling and resource management
- Telegram API – For UI delivery
- Amazon Bedrock Guardrails – For security policy enforcement
- Amazon Rekognition – For document verification
- GitHub-based continuous integration and delivery (CI/CD) pipeline – For rapid feature deployment
Key Technical Insights
The implementation revealed several crucial technical considerations. The DSPy framework was crucial in optimizing and enhancing our language model prompts. By integrating additional layers of reasoning and context awareness tools, DSPy notably improved response accuracy, consistency, and depth.
Future Enhancements
Our roadmap includes several technical improvements to enhance the system’s capabilities:
- Implementing advanced context dispatching using machine learning algorithms to improve service coordination across multiple domains
- Developing a sophisticated human-in-the-loop validation system for complex cases requiring expert oversight
- Migrating suitable components to a serverless architecture using Lambda to optimize resource utilization and costs
- Enhancing the knowledge base with advanced semantic search capabilities and automated content updates
Results
This solution, which serves hundreds of Ukrainian refugees in Denmark daily, demonstrates the potential of AWS services in creating scalable, secure, and efficient AI-powered systems for social impact.
Conclusion
Through careful architecture design and integration of complementary technologies, we’ve created a platform that effectively addresses the challenges faced by refugees while maintaining high standards of security and data protection.
About the Authors
- Taras Tsarenko is a Program Manager at Bevar Ukraine with a focus on AI-driven solutions.
- Anton Garvanko is a Senior Analytics Sales Specialist for AWS with expertise in business intelligence.
- Vitalii Bozadzhy is a Senior Developer specializing in cloud-based solutions.
- Vladyslav Horbatenko is a computer science student and Data Scientist focused on AI and LLMs.
Creating Victor: An AI-Powered Assistant for Ukrainian Refugees
Co-written by Taras Tsarenko, Vitalii Bozadzhy, and Vladyslav Horbatenko
In an era where technology meets humanitarian needs, Bevar Ukraine is paving the way with its groundbreaking AI-powered virtual assistant, Victor. Launched to support the pressing needs of Ukrainian refugees integrating into Danish society, Victor is an embodiment of innovative thinking aimed at social impact. This post explores the technical foundations of this initiative, focusing on how AWS services have enabled a scalable and secure solution.
Background and Challenges
Refugees face numerous hurdles while integrating into new societies, from accessing public services to navigating complex legal systems. Traditional methods, heavily reliant on human social workers, often fall short due to scalability issues and language barriers. Recognizing these challenges, Bevar Ukraine has developed Victor as a continuous, AI-driven solution that meets high service standards and provides timely assistance.
Solution Overview
Victor’s architecture is built on several AWS services, intended for reliability, security, and efficiency. Our volunteer software developers managed to design and deploy this system rapidly.
Key Components:
- Amazon EC2: Serves as the primary computing resource, employing Spot Instances to optimize costs.
- Amazon S3: Provides secure storage for conversation logs and essential documents.
- Amazon Bedrock: Powers the core natural language processing functionalities.
- Amazon DynamoDB: Facilitates real-time data access and session management for quick responses even during peak loads.
After careful evaluation, we concluded that Anthropic’s Claude 3.5 large language model (LLM) was the best fit for our needs due to its advanced dialogue capabilities and human-like tone. This choice enables Victor to deliver natural, engaging responses.
For multilingual support, we harnessed Amazon Titan Embeddings G1, which excels in generating high-quality vector representations of text, making semantic search efficient and effective. This functionality allows Victor to retrieve accurate information in various languages promptly.
Tips and Recommendations
Our experience in constructing Victor yielded valuable insights that can serve as guidelines for other organizations aspiring to create AI-powered solutions:
- Test Multiple Models: Use the Amazon Bedrock playground to compare LLMs and discover the most effective for your use case.
- Experiment with Prompts: Tailor your prompts and settings to enhance the quality of responses.
- Monitor Costs: Establish budgetary controls and monitoring in AWS to stay within financial constraints.
- Select Appropriate Embedding Models: Choose models that support the necessary languages and optimize settings based on your application’s requirements.
- Implement Guardrails: Use Amazon Bedrock Guardrails to enforce safety protocols, preventing the model from leaking sensitive information.
Integration and Enhancement Layer
To bolster the core AWS infrastructure further, we integrated several supporting technologies:
- Pinecone Vector Database: For efficient storage and retrieval of semantic embeddings.
- DSPy Framework: For structured prompt engineering with Claude 3.5.
- EasyWeek and Telegram API: For appointment scheduling and user interface delivery.
- Amazon Rekognition: For document verification.
- GitHub-based CI/CD Pipeline: For rapid feature deployment.
Key Technical Insights
Our implementation revealed critical insights into the design and architecture:
- Optimizing Language Model Prompts: The DSPy framework significantly improved the accuracy, consistency, and depth of responses.
- GDPR Compliance: We prioritized data minimization and secure storage, along with clear user consent mechanisms.
- Cost Optimization: Leveraging EC2 Spot Instances and API request throttling led to significant savings without diminishing performance.
Future Enhancements
Looking ahead, our roadmap for Victor includes several exciting improvements:
- Advanced context dispatching using machine learning for better service coordination across domains.
- A human-in-the-loop validation system for complex queries requiring expert input.
- Migration to a serverless architecture with AWS Lambda to optimize resource consumption.
- Enhancements to the knowledge base for improved semantic search capabilities.
Results
Victor has become an essential resource, providing automated assistance to hundreds of Ukrainian refugees in Denmark every day. This solution has saved thousands of volunteer hours, allowing staff to focus on more complex issues. For refugees, Victor is a lifeline, delivering swift answers to pressing questions regarding public services and more.
Conclusion
Through thoughtful design and technological integration, we have crafted a platform that effectively addresses the challenges faced by refugees while maintaining high standards of data protection and security. The success of this initiative provides a replicable model for similar solutions in other humanitarian contexts, harnessing the power of robust cloud infrastructure and innovative system design to foster social impact.
About the Authors
- Taras Tsarenko: Program Manager at Bevar Ukraine, specializing in AI-driven solutions and data engineering.
- Anton Garvanko: Senior Analytics Sales Specialist at AWS, with a passion for bridging finance and IT.
- Vitalii Bozadzhy: Senior Developer with expertise in cloud-based solutions and scalable architectures.
- Vladyslav Horbatenko: Data Scientist focused on AI innovations, particularly in large language models.
Together, we aim to lead transformative initiatives that utilize AI for the greater good, supporting those in need across the globe.