Concept guide · RAG foundations
What is conversational RAG?
A slower explanation for when the short interview answer is not enough. Start with the intuition, then connect it to the technical idea and the way an interviewer expects you to reason about it.
ML area
Generative AI → Conversational RAG
Interview format
Core · Scenario
Question
GENAI-055
Start with the idea
In simple terms
Conversational RAG uses earlier turns to understand a follow-up, but it should retrieve evidence for the resolved question rather than blindly sending the whole chat history every time.
Make it concrete
After asking “What is our parental leave?” a user asks “Does it apply during probation?” The system should resolve “it” to the parental-leave policy, then search for the probation rule.
Build the answer
How to approach it
- 1Resolve the follow-up into a standalone retrieval intent.
- 2Retrieve only the evidence needed for that intent with current permissions.
- 3Control history growth and test pronouns, topic shifts, and stale references.
Go one level deeper
The technical explanation
Keep the visible conversation, a retrieval query, and durable memory as separate representations. Resolve references and missing context, then apply the same permissions and freshness rules as a single-turn query. Summarise or trim history as it grows, and test follow-ups that change topic, refer to old turns, or attempt to inherit another user’s context.
Say this in the interview
Conversational RAG rewrites or interprets a follow-up question using conversation context, retrieves evidence for the resolved query, and answers with the relevant history. Keep the retrieval query separate from the visible response, control history growth, and avoid carrying irrelevant or unauthorised context between turns.
Keep preparing