Architecture scenarioMedium~10 minPrompting & Structured Output / 8 of 20

Interview question

How would you design a useful JSON schema for model output?

Designs a narrow, typed, provider-supported schema that communicates the real downstream contract.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I start from what the consumer needs and use specific types, required fields, enums for closed choices, and clear descriptions. I keep the structure shallow, disallow unexpected fields when supported, and provide an explicit representation for unknown or insufficient evidence. The schema should be stable application vocabulary, not a dump of database entities.

Full answer below

Practice focus

Practical schema design rather than treating structured output as an arbitrary serialization target.

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
The best schema is usually the smallest one that captures the downstream decision without asking the model to reproduce an internal domain model.
I start from what the consumer needs and use specific types, required fields, enums for closed choices, and clear descriptions. I keep the structure shallow, disallow unexpected fields when supported, and provide an explicit representation for unknown or insufficient evidence. The schema should be stable application vocabulary, not a dump of database entities.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
For ticket routing, I might return category, confidence band, evidence text, and a needsReview flag. I would not ask for a customer object with fields the task never uses. I check the provider’s supported JSON Schema subset because constraints such as recursion, numerical ranges, string lengths, or composition may differ or be unsupported.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I separate transport validation from domain validation. Provider-constrained decoding can enforce shape, while application code checks allowed transitions, cross-field consistency, current identifiers, and source evidence. I account for schema compilation latency and token overhead when deciding how much structure belongs in one response. The contract is versioned, and refusals or incomplete generations are represented outside the normal success payload rather than forced into domain fields.

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

Mirroring a large internal domain model.
Using free-form strings where a small enum represents the contract.
Making every field optional because the model may omit it.
Assuming unsupported schema constraints are enforced.

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

Details

The schema exposes fields the caller is not authorized to set.
There is no representation for unknown or not-applicable values.
Schema validity is treated as domain validity.

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.

When should additional properties be rejected?
How would you represent insufficient evidence?
What would you validate outside the schema?

Continue the section

Next in Prompting & Structured Output: Schema-valid but wrong

Step 9 of 20

Why can schema-valid model output still be wrong?

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

Architecture scenarioMedium~10 minPrompting & Structured Output / 8 of 20

Interview question

How would you design a useful JSON schema for model output?

Designs a narrow, typed, provider-supported schema that communicates the real downstream contract.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Full answer below

Practice focus

Practical schema design rather than treating structured output as an arbitrary serialization target.

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
The best schema is usually the smallest one that captures the downstream decision without asking the model to reproduce an internal domain model.
I start from what the consumer needs and use specific types, required fields, enums for closed choices, and clear descriptions. I keep the structure shallow, disallow unexpected fields when supported, and provide an explicit representation for unknown or insufficient evidence. The schema should be stable application vocabulary, not a dump of database entities.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
For ticket routing, I might return category, confidence band, evidence text, and a needsReview flag. I would not ask for a customer object with fields the task never uses. I check the provider’s supported JSON Schema subset because constraints such as recursion, numerical ranges, string lengths, or composition may differ or be unsupported.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I separate transport validation from domain validation. Provider-constrained decoding can enforce shape, while application code checks allowed transitions, cross-field consistency, current identifiers, and source evidence. I account for schema compilation latency and token overhead when deciding how much structure belongs in one response. The contract is versioned, and refusals or incomplete generations are represented outside the normal success payload rather than forced into domain fields.

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

Mirroring a large internal domain model.
Using free-form strings where a small enum represents the contract.
Making every field optional because the model may omit it.
Assuming unsupported schema constraints are enforced.

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

Details

The schema exposes fields the caller is not authorized to set.
There is no representation for unknown or not-applicable values.
Schema validity is treated as domain validity.

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.

When should additional properties be rejected?
How would you represent insufficient evidence?
What would you validate outside the schema?

Continue the section

Next in Prompting & Structured Output: Schema-valid but wrong

Step 9 of 20

Why can schema-valid model output still be wrong?

How would you design a useful JSON schema for model output?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

How would you design a useful JSON schema for model output?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section