A Singapore Government Agency Website
How to identify
Tool to automate report generation of sensitive security incidents for higher management
Report generation has never been this easy.
Government officers drafting sensitive reports are unable to leverage on LLMs to automate the task due to operational security. This results in extensive manual work and is time consuming.
Our tool identifies and redacts any sensitive PII data, before feeding the data into LLMs. It then un-redacts the PII data from the LLM output, ensuring the content is still coherent.
Users can customize the kind of data considered as PII, relative to their own use cases. They can also customize the LLM prompt to tailor the response. Lastly, there is a manual review in place before the data is fed into LLM as a sanity check.
Testers found it useful to be able to customize the kind of PII. For example, from a cybersecurity POV, they can customize it to identify URLs and IP Addresses.
They also found it helpful to be able to write their own LLM prompts so that different outputs can be generated. They were also pleasantly surprised at how well the tool works and the time saved.
Currently, if the user is not satisfied with the redaction, the tool terminates before feeding the data into the LLM.
We are looking to build a feature to add ad-hoc PII entities after the manual review. This way, the user can add his own redaction during the manual review before feeding it into the LLM.