Table 1 Details of the agent definition in the DisasTeller framework

From: Integration of large vision language models for efficient post-disaster damage assessment and reporting

Agents

Toolset

Task prompt input

Expert team

Image interpretation tools, local file search tool and map annotation tool

‘1. Generate descriptions for post-disaster images using the image interpretation tools; 2. Get post-disaster grading with their location names using the local file search tool; 3. Send the concise summary of disaster grades to the map annotation tool for the alert map generation; 4. Generate a report with the given format:…’

Alerts team

Global image interpretation tool, web search tool

‘Analyse the output from the Expert team, generate a comprehensive social media news that describes the damaged events for regional areas, analyse potentially dangerous areas according to the global map using the global image interpretation tool.’

Emergency team

Image interpretation tools, web search tool

‘Analyse the outputs from the Expert team and Alerts team. Then, ensure the number and locations of emergent shelters and hospitals using image interpretation tools and a web search tool to take a reference from other areas’ historical disaster data. Also, confirm which place needs immediate transportation recovery. Finally, generate a report with the given format:…’

Assignment team

Global image interpretation tool, web search tool

‘Analyse the outputs from the Expert team, Alerts team and Emergency team. 1. Generate a report for the emergency service team human resource allocation, search the internet to take a reference from other areas’ historical disaster data. The report format:… 2. Generate a report to inform the public communities, with the given format:… 3. Generate a reconstruction plan with financial budgets, aiming to submit it to the government and policymakers, with the given format:…’