Qarma AI lets you ask questions about your Qarma data in plain language, directly from the dashboards in Qarma. This article explains how it works, which data it can see, which AI models it uses, and how your data is protected.
What is Qarma AI?
Qarma AI is an AI assistant built into the dashboards in Qarma. You type a question, for example "Which suppliers had the highest defect rate last quarter?", and Qarma AI finds the relevant data, runs the query and answers with a table or chart and a short written summary. You can ask follow-up questions in the same chat to drill down.
How does it work?
- You ask a question in the chat.
- Qarma AI works out which area of Qarma's data model the question is about, for example inspections, defects or audits.
- It builds a query in Qarma's data model, the same model that powers the standard dashboards. For this step the AI model only receives your question and the names and descriptions of the available fields. It does not receive any of your records.
- The query runs against Qarma's analytics database under your own Qarma user, so it only returns the rows you are allowed to see.
- To write the summary, the query result is sent to the AI model together with your question. The answer is shown in the chat with the underlying data.
Qarma AI only reads data. It cannot create, change or delete anything in Qarma.
Which data does Qarma AI use?
Exactly the data you can already see in Qarma. Access is restricted in the same way as on the Dashboard: if you cannot see an inspection, audit or supplier in Qarma, Qarma AI cannot see it either when you ask a question.
So you do not need to worry about giving users access to Qarma AI - they will not be able to see more data than what they can already show by navigating the system already.
Which AI models does Qarma AI use?
Qarma AI uses Claude Sonnet 5.5 from Anthropic, run in AWS Bedrock. Is any data shared with Anthropic?
No. The Claude models run inside AWS Bedrock. Your questions and data are processed in AWS and are not sent to Anthropic.
Your questions, the field descriptions and the query results sent to the model are not used to train any AI model.
Where is the data processed?
Qarma AI's analytics platform is hosted in the EU. It is mentioned in Qarma's Data Processor Agreement that all customers are covered by.
How does Qarma AI relate to the EU AI Act?
Qarma has assessed its AI features against the EU AI Act (Regulation (EU) 2024/1689). None of them are prohibited or high-risk. Qarma AI falls under the transparency rules for systems people interact with, so the chat is clearly presented as AI. Qarma also has an internal AI use policy and runs regular AI training for its staff.
Can the answers be wrong?
Yes. Like any AI system, Qarma AI can misread a question or pick the wrong field. Each answer shows the data it is based on, so check the table or chart before you act on an answer or share it, and rephrase the question if the result does not look right.