RAG Interview Questions
Dashboard
0%
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
51 small wins to finish your pathNext question →

Q2EasyConcept

Walk through the two phases of a RAG pipeline: indexing and querying.

30-second answerSay your answer out loud first, then reveal.
The two RAG phases: offline indexing parses, cleans and chunks sources into a vector index and a BM25 keyword index; online querying rewrites the question, retrieves from both indexes, reranks and lets the LLM answer with citations.

Indexing details

  • Loading / parsing: extract text from PDFs, HTML, DOCX, slides; handle tables and images (Q15).
  • Cleaning: remove boilerplate (headers, footers, navigation), fix encoding, deduplicate.
  • Chunking: split into retrievable units (Q4).
  • Enrichment: attach metadata (source, title, date, section, access permissions); optionally add contextual summaries (Q22).
  • Embedding + storage: vectors in a vector DB; often also a BM25 index for keyword search.

Querying details

  • Query processing: rewrite follow-up questions, expand acronyms, decompose complex questions (Q16).
  • Retrieval: dense (vector), sparse (BM25) or hybrid; metadata filters.
  • Post-retrieval: rerank, deduplicate, compress, order the chunks.
  • Generation: prompt with instructions to answer only from the context and to cite sources.
Interview tip. Most RAG quality problems originate in indexing (bad parsing, bad chunks), not in the LLM. Saying this out loud signals real experience.
PreviousWhat is RAG, and why do we need it?

You understood something today that you didn't yesterday.