Comparison questionMedium~10 minRAG & Vector Search / 2 of 18

Interview question

When would you use RAG instead of a database query or tool call?

Chooses retrieval for ambiguous knowledge search while keeping authoritative structured operations behind deterministic services.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I use RAG when the user asks an open-ended question over unstructured or semi-structured knowledge such as manuals, policies, or incident notes. I use a database query or tool when the answer depends on exact current state, a calculation, or an operation such as an account balance or order cancellation. The deciding factor is the source and precision the task requires, not whether an LLM is available.

Full answer below

Practice focus

Judgment about unstructured search, source authority, precision, and side effects.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
RAG is a search-and-context pattern, not a replacement for authoritative application queries.
I use RAG when the user asks an open-ended question over unstructured or semi-structured knowledge such as manuals, policies, or incident notes. I use a database query or tool when the answer depends on exact current state, a calculation, or an operation such as an account balance or order cancellation. The deciding factor is the source and precision the task requires, not whether an LLM is available.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
A question like how refunds generally work can retrieve policy text, but whether this customer's order is refundable should call the order service under that user's authorization. A combined flow can retrieve the policy and fetch the order state, then let the model explain the result. I keep those sources labeled so generated prose cannot blur a general document with an authoritative record.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
Vector search is approximate and document indexes can lag behind source systems, so it is a poor authority for transactional truth. Tool and query results also need typed validation, timeout handling, and provenance before the model sees them. I design an explicit evidence plan that states which source can answer each part of the question. If sources conflict, deterministic policy decides precedence and the response exposes uncertainty rather than allowing the model to reconcile authority silently.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Putting relational rows into a vector index for every exact lookup.
Using RAG to perform a state-changing action.
Letting the model decide which source is authoritative.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

The index is treated as the system of record.
Approximate similarity controls money or permissions.
Structured and retrieved evidence are indistinguishable in traces.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

How would you combine retrieval with tool use?
What if policy text and live state disagree?
Which source should be cited to the user?

Continue the section

Next in RAG & Vector Search: RAG ingestion pipeline

Step 3 of 18

How would you design an ingestion and indexing pipeline for RAG?

Practice route

Read core answer Review checklist Follow-up prompts Track progress

Finished this drill?

Practice freely. Sign in only when you want to save completion, confidence, notes, or bookmarks.

New to Aporeon? Create account

Drill details

Human editorial direction

Aporeon is shaped by Aleksandar Tomovski, a software developer with experience on both sides of technical interviews. Content is reviewed for accuracy, natural spoken delivery, useful depth, and honest trade-offs.

About the creator Editorial standards

Was this drill useful?

Help improve the interview library.

Found this answer useful?

Say thanks with a standalone, one-time $5 contribution. No account required.

No subscription or perks.Choose another amount

Opening drill

Comparison questionMedium~10 minRAG & Vector Search / 2 of 18

Interview question

When would you use RAG instead of a database query or tool call?

Chooses retrieval for ambiguous knowledge search while keeping authoritative structured operations behind deterministic services.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Full answer below

Practice focus

Judgment about unstructured search, source authority, precision, and side effects.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
RAG is a search-and-context pattern, not a replacement for authoritative application queries.
I use RAG when the user asks an open-ended question over unstructured or semi-structured knowledge such as manuals, policies, or incident notes. I use a database query or tool when the answer depends on exact current state, a calculation, or an operation such as an account balance or order cancellation. The deciding factor is the source and precision the task requires, not whether an LLM is available.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
A question like how refunds generally work can retrieve policy text, but whether this customer's order is refundable should call the order service under that user's authorization. A combined flow can retrieve the policy and fetch the order state, then let the model explain the result. I keep those sources labeled so generated prose cannot blur a general document with an authoritative record.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
Vector search is approximate and document indexes can lag behind source systems, so it is a poor authority for transactional truth. Tool and query results also need typed validation, timeout handling, and provenance before the model sees them. I design an explicit evidence plan that states which source can answer each part of the question. If sources conflict, deterministic policy decides precedence and the response exposes uncertainty rather than allowing the model to reconcile authority silently.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Putting relational rows into a vector index for every exact lookup.
Using RAG to perform a state-changing action.
Letting the model decide which source is authoritative.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

The index is treated as the system of record.
Approximate similarity controls money or permissions.
Structured and retrieved evidence are indistinguishable in traces.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

How would you combine retrieval with tool use?
What if policy text and live state disagree?
Which source should be cited to the user?

Continue the section

Next in RAG & Vector Search: RAG ingestion pipeline

Step 3 of 18

How would you design an ingestion and indexing pipeline for RAG?

When would you use RAG instead of a database query or tool call?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

When would you use RAG instead of a database query or tool call?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section