Knowledge Assistant (RAG)
Knowledge Assistant is an AI product based on RAG (Retrieval-Augmented Generation) that combines search across an organization's knowledge base with answer generation based on an LLM.
The product exists in two variants: for internal users (employees) and for external users (customers).
Purpose
The Knowledge Assistant addresses fast access to an organization's knowledge. Instead of manually searching through documents, the user asks a question in natural language and receives a generated answer with links to sources.
RAG Architecture
A typical RAG system architecture includes the following components:
- Document ingestion — connecting to data sources
- Chunking — splitting documents into fragments
- Embedding — vectorizing the fragments
- Vector store — indexing and storing the embeddings
- Retrieval — finding relevant fragments for a query
- Generation — forming an answer based on the retrieved fragments via an LLM
- Answer to the user — an answer with citations and links to sources
Key Aspects
When developing and operating the Knowledge Assistant, you need to consider:
- Data quality — answers are only as good as the source documents
- Retrieval relevance — retrieval accuracy directly affects answer quality
- Hallucination control — an LLM may generate inaccurate information
- Access control — a user should see only the documents they have access to
- Data freshness — documents require regular updating in the index
Two Product Variants
The Knowledge Assistant is implemented in two variants with different requirements:
| Aspect | Internal | External |
|---|---|---|
| Users | Employees | Customers |
| Data | Internal documents, policies | Public documentation, FAQs |
| Error tolerance | Medium | Low |
| Access control | By roles and departments | By products and plans |
| Escalation | To an expert | To a support agent |