Product Knowledge Grounding

When users ask questions about how OrviQ works, what specific statuses mean, or how to perform a task, OrviQ Assistant retrieves content directly from the Product Knowledge Corpus.

---

Retrieval Mechanics

1. Intent Classification: The system identifies that the question is a Product Help inquiry (e.g., "How does RCSA work?" or "What is the difference between Subset and Superset?").
2. Vector Similarity Search: The Assistant searches the published documentation corpus using pgvector cosine similarity.
3. Relevance Thresholding: Only excerpts exceeding a strict similarity threshold are passed to the model. If no relevant documentation exists, the Assistant transparently explains that no official documentation was found rather than inventing behavior.
4. Trusted Citation Construction: Answers cite the exact documentation article and section anchor on https://docs.orviq.io.