Important information is often scattered across documents, shared drives, internal tools and the memory of a few colleagues.
We build assistants that let authorised employees ask questions in natural language and get answers drawn from approved company sources, with links to the documents they came from.
What we deliver
- Source selection and document preparation
- Indexing with semantic and keyword search
- Connections to approved repositories such as SharePoint or Google Drive
- Answers with source citations
- Permission-aware retrieval
- Scheduled re-indexing as documents change
- An evaluation set and quality reports
- A clear fallback when no reliable answer exists
Technology & implementation
Most knowledge assistants use retrieval-augmented generation (RAG). Documents are split into passages and indexed, usually with embeddings (numerical representations of meaning) stored in a vector database, combined with keyword search for names, codes and numbers. When someone asks a question, the most relevant passages are retrieved and the model writes its answer from them, citing each source.
Permissions apply at retrieval: a passage from a document the user cannot open is not retrieved for them. The index is refreshed when documents change, and outdated or duplicate files are excluded, because an assistant can only be as good as what it finds.
Source references let users check which documents support an answer, but they do not guarantee that the answer is correct. Retrieval quality, permissions and answer accuracy still need to be tested: quality is measured against a set of real questions with expected answers, the assistant is tested on saying when it does not know, and that test set keeps running as the documents change.
Example use
An illustrative scenario, not a description of a completed project.
An employee asks how to complete an internal procedure. The assistant locates the relevant approved documentation, summarises the steps and links to the original sources.