A naukri.com initiative
Dev
1w
230
Image Credit: Dev
Building "Yuh Hear Dem": A Parliamentary AI with Google's ADK and a Lesson in Agentic Design
- The project 'Yuh Hear Dem' was developed as a response to the challenge of making parliamentary data accessible to citizens, utilizing Google's Agent Development Kit (ADK) Hackathon for development.
- The system, designed for civic engagement in Barbados, transforms unstructured parliamentary data into a queryable format, allowing citizens to ask natural language questions and receive verified answers.
- It utilizes a sophisticated Retrieval-Augmented Generation (RAG) pipeline, combining vector search and a knowledge graph for accuracy and precision in responses.
- The technical architecture includes a three-stage pipeline for data processing and retrieval, starting with ingesting, cleaning, and extracting raw transcripts to build a Knowledge Graph.
- The system implements hybrid retrieval with GraphRAG, which enhances search capabilities for both broad and specific queries, ensuring information accuracy.
- The architecture evolved from a multi-agent system to a single intelligent agent solution due to state management complexities, ensuring reliability and ease of debugging.
- The user experience focuses on making AI accessible through progressive disclosure, visual learning with interactive graphs, and contextual guidance for natural exploration paths.
- Key lessons learned include the effectiveness of hybrid RAG, the robustness of ADK's single-agent model, the importance of pragmatic architecture, and the necessity of human-centered design for user accessibility.
- Future plans involve expanding data sources, potentially returning to a multi-agent architecture, and continuing to enhance democratic transparency through technology.
- The 'Yuh Hear Dem' project aims to empower citizens by providing a tool to engage with parliamentary data effectively.
- The project website, GitHub repository, and hackathon submission links are provided for further exploration.
Read Full Article
13 Likes
For uninterrupted reading, download the app