Zhiva — Intelligent Company Knowledge Base
The system answers questions from employees and external audiences based on company materials. It generates responses from the knowledge base and backs them with source links.
The task
Turn a company’s materials into an accessible knowledge base that staff and external audiences can query in natural language.
Search and answer generation
The system uses the company’s own material to find information and compose answers. Each answer carries a link to the specific clause of the document, so the user can verify the source.
If the knowledge base does not contain the answer, the system reports that the information is missing.
Deployment and quality assessment
Internal and external access, a Telegram interface, and deployment within the company’s own infrastructure, with no documents leaving it, are all supported.
In an internal comparison on 56 questions, 80% of answers were correct, versus 37% for a baseline search over text fragments. This result applies to the specific test set used in that evaluation.