Corporate ownership charts take a long time to manually read over and understand. These diagrams can contain lots of information such as what companies own others, locations of companies, ownership percentages, etc. Oftentimes, these diagrams are saved as some form of image file. The problem we want to solve is that a large amount of time is being taken to manually analyze all of those diagrams and find the desired information. To solve this, we will create an automated way to extract the data from these diagrams, so that Kingland can automatically load data into their application.
Our project idea is to create a multi-agent workflow in which we have multiple sub-agents with various tasks that all work together to extract the information from these corporate family tree diagrams from various file formats (pdfs, pngs, pptxs, xlsx, etc.) and produce a JSON output with all the data from the diagrams.
The multi-agent workflow handles communication between a primary agent and multiple subagents. Subagents will be calculated proportional to the size of each given corporate ownership chart fed into the system. Each subagent will receive a broken-down version of the image and will extract the relation information. Once extracted, the information will be fed back to the primary agent, where it will be compared against in-instance memory to determine the validation percentage and confidence score. Once all subagents have broken down their images into information, the primary agent will rebuild the corporate family tree into JSON format for the Kingland user to use for their proprietary software.
