Page 677 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Key Aspects of List main elements of the proposed solution, including the role of AI.
Solution 1.Large Language Models (LLMs): cities 4.8: Smart home/
Handle multilingual citizen inquiries with 24/7 responsiveness; automatically
parse complaint tickets to extract key information (location/issue type) and
generate resolution recommendations.
2.Multimodal Large Models:
Simultaneously analyze unstructured data uploaded by citizens (e.g., "text
descriptions + on-site photos") to accurately identify complex, hybrid issues.
3.Visual Large Models:
Monitor urban issues in real-time via cameras and promptly provide early
warning alerts.
4.RAG (Retrieval-Augmented Generation):
Dynamically retrieve the latest policy documents to ensure AI responses
remain compliant; match historical similar cases to provide actionable reso-
lution templates.

Technology Relevant technical keywords
Keywords LLMs, Multimodal large model, RAG, Fine-tuning

Data Availability Public / Private (specify sources or repositories if available)
Private - Urban data

Metadata (Type Specify data types (e.g., text, audio, visual)
of Data) Structured and unstructured data including natural language text, policies,
reports, video and image

Model Training Briefly describe AI/ML technologies and approaches used 1.Pretrain with
and Fine-Tuning Urban data
2.Supervised Fine-Tuning(SFT)

Testbeds or Pilot Provide links or references to deployment examples
Deployments It is private.

Code reposito- Provide github links (if available)
ries It is private.

2 Use Case Description

2�1 Description

Yuanjing Integrated Urban Management Platform is based on advanced large-model
technology, aiming to promote sustainable development in cities worldwide, enhance urban
governance efficiency, optimize public service responsiveness, and create a smarter and more
efficient living and working environment for citizens. By leveraging language large models,
vision large models, and multimodal large models, the platform has implemented functions
such as intelligent Q&A for public services, rapid dispatch of requests, precise detection of
urban events, and real-time early warning and scheduling. With its efficient multi-mode data
processing capabilities, the platform also supports the expansion of large-scale application

641