Using AI to process confidential information: building a secure solution with Atostek’s language model

Cloud-based AI solutions may not allow customers to utilize confidential data. Atostek offers a local language model run from the company’s own data center, which be used for purposes such as compressing large volumes of patient records into an easily usable format.

Atostek’s language model currently supports over one hundred languages. Our local language model allows our customers to process confidential and sensitive information in a secure way. The language model may be applied to any type of data.

For example, health records and patient records tend to include a lot of free-form text. Previously, the lack of structural data has prevented patient records from being used for applications such as calculating health risks. However, with the help of a language model that is able to process natural language, this free-text content can be transformed into an easily readable format, and long texts can be used to acquire initial data for risk calculation.

Atostek’s servers are located in the company’s own facilities. The server has already been used for generating summaries similar to the ones mentioned above. In order to produce relevant results, language models must be carefully trained. For this reason, Atostek has further trained its language model to produce summaries of materials containing tens of thousands of example files.