Cogniss white-label AI platform
Helping organizations make their knowledge accessible to people and AI systems while retaining control over data, governance, and sovereign AI deployment within the EU.
Challenges
There was no way to see inside a running workflow
AI workflows could involve multiple agents, models, data sources, and asynchronous execution states, making it difficult to understand what happened, why a run failed or where a result came from.
Permissions were hard to reason about
Permissions were resolved across platform access, roles, groups, and individual users, requiring admins to see not only what someone could do, but why they had that access.
Refining a report could undo human edits
AI-generated reports needed to support substantial refinement without silently replacing work a user had already reviewed or edited.
Advanced logic had to stay usable by non-engineers
Powerful capabilities such as conditional logic, external connectors, and agent tools had to remain usable without turning the product into a visual programming environment.
The Work
Defined a shared permissions model
We defined a permissions model shared across roles, groups, and effective user access, with clear treatment of inherited and overridden permissions.
Made AI execution traceable
We designed execution and traceability patterns that expose workflow states, models, inputs, outputs, and source data instead of hiding AI activity behind a generic loading state.
Kept users in control of AI-assisted changes
We established a preview-before-apply pattern for AI-generated and AI-refined report content, allowing users to review proposed changes before accepting them. For AI-assisted data mapping, we designed controls to inspect suggestions and accept or reject individual mappings.
Built complex AI controls around familiar patterns
We designed configuration tools for AI agents operating across multiple steps of the chat and reporting pipelines. We reused existing permissions and interaction patterns wherever possible, instead of introducing separate controls and additional administration complexity.
Identified implementation gaps before further development
We audited existing reporting and external-source workflows against their specifications, testing implemented functionality and documenting discrepancies. The findings helped distinguish interface improvements from deeper technical questions, providing a basis for re-estimation and decisions about scope.
Outcomes
Testing existing reporting and external-source features exposed gaps between the specifications and the implementation, including technical questions that required investigation. The findings informed scope decisions and provided a clearer basis for re-estimating the work. In parallel, we prepared development-ready designs for agent configuration and AI-assisted workflows, reusing existing interaction patterns to keep complex capabilities manageable.