Responsible AI

Explainability

• Accuracy: An automated verification process for accuracy is executed periodically and prior to any production deployment.

• Traceability: C-Agent cites the source of data used in its responses. For business narratives, it provides a direct link to open the referenced document.

Fairness

• Representative and Diverse Data: The context was defined separately for each of C-Agent’s knowledge domains (semantic fields), ensuring coverage of diverse perspectives.

• Bias Mitigation Algorithms: By leveraging multiple agents, C-Agent delivers complementary answers to the same query, reducing the risk of systemic bias.

• Multidisciplinary Development Teams: C-Agent is developed through collaboration among experts from diverse fields, including Controllership, Data Science, Artificial Intelligence, Python Development, and Azure technologies.

Robustness

C-Agent employs Azure OpenAI content filters, complemented by a Retrieval-Augmented Generation (RAG) model with content validation for acceptance or rejection. A whitelist mechanism for user management and role-based access ensures that information is visible only to authorized users. This design enforces strict data governance and protection at the source.

Transparency

From the welcome message, C-Agent provides access to user manuals, allowing users to consult documentation at any point.

Privacy

C-Agent uses only the user’s email address for authentication and query history. It does not process or store additional personal data at any stage. By excluding personal data from training processes, privacy risks are significantly reduced.