Retrieval: treating provenance as a first-class concern
Inside an organisation an answer without provenance cannot be trusted. We put citation ahead of raw retrieval quality.
The common failure mode in a knowledge-base project is not "it cannot be found" but "it was found and nobody dares use it" — usually because the answer cannot be traced to a specific document and location.
The current direction is to make citation a fixed output of the retrieval pipeline rather than an addendum at generation time. That changes chunking strategy: a chunk has to retain enough positional information to locate itself.
This is a record of a research direction, not a statement of completed capability.
Related applications
Enterprise Knowledge & Internal Q&A
Knowledge Fabric
Enterprise DeliveryTurning documents, policies, project material and system data into enterprise capability that can be searched, cited and kept current.
AI Data Analysis & Decision Support
Insight Engine
Private PreviewConnecting business data so querying, analysis, anomaly detection, trend reading and reporting can happen in natural language.
