Portfolio Details

Abstract

This project lays the foundation for a Generative AI Retrieval-Augmented Generation (RAG) chatbot aimed at transforming how city governments deliver accessible and timely information to residents and the public. The chatbot addresses a critical gap in providing real-time support for frequently asked questions, public services, and key information such as policies, city budgets, and spending data. By offering clear and straightforward responses, the chatbot seeks to enhance public engagement and ensure reliable information is available 24/7, particularly during off-hours or periods of high demand. The chatbot prototype leverages the AWS QnAbot framework, integrating Amazon Kendra to build and maintain an up-to-date knowledge base from city government resources and Anthropic Claude 3.5 Sonnet via Amazon Bedrock to deliver conversational responses. Prompt engineering techniques were implemented to ensure the chatbot handles diverse queries effectively, providing personalized answers while maintaining clarity and relevance. Manual testing was carried out to evaluate the chatbot’s functionality, assess its error handling capabilities and the testing outcomes shows the chatbot’s effectiveness in providing reliable responses while handling diverse queries. Test cases and findings were documented to verify that the chatbot delivers the intended outcomes. The final deliverables will include documentation on integration, prompt engineering strategy, and operational guidelines, laying the groundwork for an innovative, user-friendly tool to improve public service communication.

AI Integration

• Incorporated Anthropic Claude LLM with a Retrieval-Augmented Generation (RAG) architecture to deliver accurate, contextually relevant, and timely responses to diverse user queries.
• Enhanced the chatbot's ability to support Virginia Beach residents by addressing their queries with precision and reliability.

Dynamic Data Retrieval

• Leveraged AWS tools, including Amazon Kendra, Bedrock, Lambda, and DynamoDB, to index and retrieve real-time information from the Virginia Beach knowledge base.
• Ensured the chatbot could provide up-to-date and relevant responses sourced directly from official city resources, such as policies, events, and FAQs.

Testing and Validation

• Developed a comprehensive manual testing framework to assess the chatbot's performance.
• Testing categories included:
• Response quality
• Contextual understanding
• Error handling
• User satisfaction
• Evaluated over 200 city-related queries, ensuring the chatbot delivered accurate and reliable information.

User-Centric Design

• Designed the chatbot interface in alignment with Virginia Beach City Government branding to ensure an intuitive and accessible user experience.
• Prioritized ease of use for all residents, including plans for future multilingual support to accommodate the city's diverse population.

Scalable and Secure Architecture

• Implemented AWS CloudFormation templates to automate infrastructure deployment, ensuring scalability and security.
• Integrated robust data protection measures, including encryption and access controls, to safeguard sensitive user data and maintain compliance with public sector standards.

Operational Efficiency

• Ensured 24/7 availability of the chatbot, enabling consistent and uninterrupted access to information for residents at any time, including during off-hours and peak usage periods.