VeriGov

VeriGov is an AI tool which analyses government provided datasets and claims and summarises them. An LLM-Agent is used to webscrape news websites and other sources which provide public records. This data is then assimilated into general, verbose remarks which are quantified and compared against the government records. A mismatch in the relative values of both the records may indicate possible malicious alteration of the government data. VeriGov also triggers attention when extreme values are reported in either of the records to alert possible neglection or mismanagement of responsibility. Insufficient data also raises an alert.

Currently, both the officially provided data and the collected public records are not real datasets, rather they were artificially generated for the purpose of this project. The government does not provide datasets with standardised formatting, hence webscraping or using their APIs would not have been effective. The data used still resembles what would have been the output of an API request to their website, had the resulting data been standardised. The public records were not developed using an LLM-Agent as that would require more computing power and a self trained LLM model to guarantee accuracy. Since the aim of this project was to illustrate detection of fraud, the idea of data retrieval is only conceptualised and not implemented.

AI is prone to hallucinations or misjudgments. This may lead to seemingly accurate results which are completely false in reality. Therefore, the use of VeriGov should be restricted to usage as a tool to aid humans and not as concrete evidence or testament These alerts are only prompts to further investigation and not accusations.

Results

Here are the findings from the analysis. The table displays government and implied public records based on the comments. Click on the entries in the 'Public Records' column to view the details indicated by the remarks. Please note that all interpretations are compared to a control value within each dataset, making them dynamic. Therefore, the value of 'Sufficient' may vary across different datasets. Additionally, a 'Credibility Score' is calculated based on discrepancies and significant alerts collected. A higher score indicates a lower necessity for additional investigation.

Title

Date	Government Records	Public Records (Remarks)

Mismatch
This shows possible fraudulent activity due to mismatched government and public records.

Attention
Extreme values being reported in both records may be of concern and should be addressed.

Insufficient Data
The lack of data in either of the records prevents analysis. Please check availability of data.

Summary
Dataset Credibility Score: 4/10

Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Diam phasellus vestibulum lorem sed risus ultricies tristique. A arcu cursus vitae congue mauris rhoncus aenean vel. Sed cras ornare arcu dui vivamus arcu felis bibendum. Nulla pharetra diam sit amet nisl. Diam phasellus vestibulum lorem sed risus ultricies tristique. A arcu cursus vitae congue mauris rhoncus aenean vel.