GovGPT, powered by Spesland, aims to revolutionize urban management and governance through cutting-edge AI-driven technologies. The company's mission is to catalyze urban excellence by optimizing resource utilization, reducing environmental impact, and enhancing public services with innovative AI solutions.
City managers face significant challenges such as aging infrastructure, unsatisfactory city infrastructure and services, and a lack of effective utilization of IoT and AI in city management. These challenges lead to lengthy and costly processes in public tenders, inefficient government officials, limited AI applications in critical infrastructure, and poor decision-making in areas like budgeting, city planning, and traffic control.
Initial Approach: To tackle the complexities of urban management, I first split the tender data into text chunks. Using a custom BERT model, these chunks were embedded into high-dimensional vectors, capturing the nuanced aspects of urban planning and governance. These vectors were then efficiently stored in Milvus, a vector database, setting the foundation for our advanced analysis.
Incorporating RAG: Central to our strategy was the use of Retrieval Augmented Generation (RAG). This approach allowed us to dynamically retrieve relevant information from our vector database during the generation process. By doing this, we ensured that our AI model was not only generating content based on its training but also utilizing specific, contextually relevant data from our urban management database.
Fine-Tuning Outputs: The final, crucial step involved fine-tuning the outputs of the RAG process. This post-processing refinement ensured that the generated summaries and recommendations were not only relevant but also precise and actionable for city managers. By tweaking the RAG's output, we were able to tailor the information to the specific needs and challenges of urban governance.
Current Progress and Future Directions: The work we've done so far has laid a strong foundation for harnessing AI in urban management. While our current results have demonstrated the potential of combining advanced AI techniques like Retrieval Augmented Generation and custom BERT models, we recognize that this is an evolving journey. Our ongoing efforts are focused on refining the accuracy and relevance of our AI-generated insights, with the aim of continuously improving their applicability to real-world urban governance challenges. This project, still in its dynamic phase, promises to evolve further, adapting and enhancing its capabilities to meet the ever-changing needs of city management.