Self-hosted RAG chatbot struggles with 3,000 internal docs
A company wants to build an internal AI knowledgebase for more than 3,000 documents, including policies, standard procedures, and general . Employees should be able to ask a chatbot instead of searching through folders by hand.
Because the documents contain sensitive data, SaaS tools are not allowed, and the system must be fully . was tested but gave inaccurate answers about 8 out of 10 times, even on simple questions.
Workspaces, custom prompts, document preparation, and chunking were already tried, but the results were still not reliable enough for employees. The needed tool must handle large document , separate workspaces or tenants, strong RAG accuracy, and reasonable setup and work for a company.
Key points
- The use case involves more than 3,000 internal company documents.
- Sensitive data makes SaaS unsuitable, so the system must be .
- returned wrong answers about 8 out of 10 times in testing.
- Custom prompts, workspaces, document preparation, and chunking did not fix the problem.
- The main are large document handling, RAG accuracy, separated workspaces, and manageable .