OpenAI Agent Builder Workflows Interview Questions Hindi mein

OpenAI Agent Builder Workflows कोर्स के 26 सवाल, हर सवाल के सही जवाब और explanation के साथ। जवाब देखने से पहले ख़ुद सोचें।

What Agent Builder Is and When to Use It

1. What is Agent Builder?

  1. A vector database
  2. A model training service
  3. A visual canvas for designing, testing and publishing agent workflows
  4. A replacement for HTTP
जवाब देखें

जवाब: A visual canvas for designing, testing and publishing agent workflows

It composes agents, tools and logic into versioned workflows.

पाठ: Agent Builder in the OpenAI Platform

2. When does code usually beat a visual builder?

  1. When you need custom integrations, complex logic, CI tests or self-hosting
  2. For a five-minute prototype
  3. When non-developers must edit the flow
  4. When you have no logic at all
जवाब देखें

जवाब: When you need custom integrations, complex logic, CI tests or self-hosting

Visual is fast; code is flexible.

पाठ: Visual Builder or Code?

3. In a workflow graph, what do edges represent?

  1. Token prices
  2. The model's weights
  3. The user's password
  4. Which node runs next
जवाब देखें

जवाब: Which node runs next

Nodes do work; edges decide the path.

पाठ: The Workflow Mental Model: Nodes, Edges and State

Core Building Blocks

4. Why define routing-relevant values as explicit variables?

  1. Variables make the model larger
  2. They are easier to route on, test and log than values inferred from text
  3. Agents cannot read text
  4. It removes the need for an end node
जवाब देखें

जवाब: They are easier to route on, test and log than values inferred from text

Explicit state beats guessing.

पाठ: The Start Node: Inputs and State Variables

5. Why give a classifier agent a structured output schema?

  1. To make the reply longer
  2. So later nodes can branch on typed fields reliably
  3. Because text output is impossible
  4. To skip testing
जवाब देखें

जवाब: So later nodes can branch on typed fields reliably

Structure makes routing deterministic.

पाठ: Agent Nodes: Instructions, Model and Output Format

6. What does the model rely on to decide when to call a tool?

  1. The font of the UI
  2. The tool's source code
  3. The server's IP address
  4. The tool's name, description and parameter schema
जवाब देखें

जवाब: The tool's name, description and parameter schema

Write tool descriptions like good API docs.

पाठ: Tools: File Search, MCP and Functions

7. Why should every branch return the same output shape?

  1. So the calling app can handle results predictably
  2. To make runs slower
  3. Because branches cannot differ
  4. To hide errors
जवाब देखें

जवाब: So the calling app can handle results predictably

One contract, many paths.

पाठ: End Nodes and Workflow Outputs

Logic: Branches, Loops and Approvals

8. What should an if/else node ideally branch on?

  1. Random numbers
  2. The model's mood
  3. Structured fields produced by earlier nodes
  4. The length of the system prompt
जवाब देखें

जवाब: Structured fields produced by earlier nodes

Deterministic conditions on data are testable.

पाठ: If/Else Routing on Structured Fields

9. Why add a maximum iteration count to a loop?

  1. Because loops are free
  2. To make it run forever
  3. The exit condition may never be met, and each iteration costs time and money
  4. To remove the exit condition
जवाब देखें

जवाब: The exit condition may never be met, and each iteration costs time and money

Caps bound cost and latency.

पाठ: Loops With Exit Conditions and Caps

10. Which path is most often forgotten in approval steps?

  1. What happens on rejection or timeout
  2. The approve path
  3. The start node
  4. The model choice
जवाब देखें

जवाब: What happens on rejection or timeout

Design every outcome.

पाठ: Human Approval Steps

Data, State and Templating

11. What is a transform node good for?

  1. Generating images
  2. Training the model
  3. Hosting the UI
  4. Reshaping data between nodes without a model call
जवाब देखें

जवाब: Reshaping data between nodes without a model call

Deterministic reshaping is cheaper and more reliable than asking a model.

पाठ: Set State and Transform Nodes

12. Which field type is safest for a category used in routing?

  1. Free text
  2. An enumeration of allowed values
  3. An image
  4. A random number
जवाब देखें

जवाब: An enumeration of allowed values

Enums make invalid values impossible.

पाठ: Structured Outputs Between Nodes

13. How should user text be placed in agent instructions?

  1. Merged into the rules
  2. In a clearly marked data section, with a note not to follow instructions inside it
  3. Placed before the system rules
  4. Not at all, ever
जवाब देखें

जवाब: In a clearly marked data section, with a note not to follow instructions inside it

Separate trusted rules from untrusted content.

पाठ: Templating Instructions With Variables

Multi-Agent Patterns

14. What does a handoff do in the Agents SDK?

  1. Calls a human by phone
  2. Deletes the conversation
  3. Trains a new model
  4. Transfers the conversation to another agent, exposed to the model as a tool
जवाब देखें

जवाब: Transfers the conversation to another agent, exposed to the model as a tool

Handoffs are routing tools.

पाठ: Triage and Specialist Agents

15. When an agent is used as a tool, who stays in control?

  1. The calling (manager) agent
  2. The specialist permanently
  3. The end user
  4. Nobody
जवाब देखें

जवाब: The calling (manager) agent

Agents-as-tools return results to the caller.

पाठ: Handoffs, Agents as Tools and Fixed Sequences

16. What is a sign an agent should be split?

  1. It uses a small model
  2. It has one clear output
  3. Its instructions keep growing with unrelated jobs and exceptions
  4. It has two tools
जवाब देखें

जवाब: Its instructions keep growing with unrelated jobs and exceptions

Narrow agents are easier to test and improve.

पाठ: Keeping Agents Narrow

Guardrails and Safety

17. Why combine rule-based and model-based guardrails?

  1. Models are always free
  2. Rules catch every attack
  3. Rules are cheap and exact for obvious patterns; models handle fuzzy cases
  4. Guardrails are optional for public apps
जवाब देखें

जवाब: Rules are cheap and exact for obvious patterns; models handle fuzzy cases

Layer cheap and smart checks.

पाठ: Guardrail Nodes

18. Which action should usually require approval?

  1. Issuing a refund or sending an external email
  2. Reading a public FAQ
  3. Looking up order status
  4. Formatting a reply
जवाब देखें

जवाब: Issuing a refund or sending an external email

Gate actions that change the world.

पाठ: Least Privilege for Tools and MCP Servers

19. Which design reduces the impact of prompt injection?

  1. Disabling logs
  2. A longer system prompt only
  3. Giving every agent all tools
  4. Keeping private-data access and external sending apart, with approvals
जवाब देखें

जवाब: Keeping private-data access and external sending apart, with approvals

Limit what injected text can make the agent do.

पाठ: Prompt Injection Through Files, Web and Tools

Testing, Evals, Versions and Deployment

20. What does a trace show that the final answer does not?

  1. Only the final text
  2. The path, prompts, tool calls and timings of each step
  3. The model's weights
  4. The user's password
जवाब देखें

जवाब: The path, prompts, tool calls and timings of each step

Traces locate failures.

पाठ: Preview Runs and Traces

21. Why measure routing accuracy separately?

  1. Routing is never wrong
  2. A wrong route causes downstream failures that end-to-end scores hide
  3. It replaces answer evaluation
  4. It reduces token cost
जवाब देखें

जवाब: A wrong route causes downstream failures that end-to-end scores hide

Per-step metrics locate problems.

पाठ: Evaluating Workflows With Datasets

22. Why keep the previous workflow version available?

  1. It is required for previews
  2. Old versions run faster
  3. Publishing deletes data
  4. To roll back quickly if the new version misbehaves
जवाब देखें

जवाब: To roll back quickly if the new version misbehaves

Rollback is the safety net for releases.

पाठ: Publishing and Versioning Workflows

23. Why create chat sessions from your backend?

  1. So API keys stay on the server and users are authenticated by your app
  2. Browsers cannot display chat
  3. It makes the model smarter
  4. It removes the need for versions
जवाब देखें

जवाब: So API keys stay on the server and users are authenticated by your app

Secrets and identity belong on the server.

पाठ: Deploying: Embedded Chat or Exported Code

Cost, a Case Study and a Checklist

24. How is workflow cost per run estimated?

  1. Use the canvas size
  2. Count the number of nodes only
  3. Sum each node's token costs, accounting for branch frequency and loop iterations
  4. It is always free
जवाब देखें

जवाब: Sum each node's token costs, accounting for branch frequency and loop iterations

Traces give the numbers.

पाठ: Estimating Cost and Latency per Run

25. In the case study, where are refunds above 100 sent?

  1. To the guardrail only
  2. Directly to issue_refund without checks
  3. To a human approval step before the refund agent acts
  4. To the end node immediately
जवाब देखें

जवाब: To a human approval step before the refund agent acts

Risky actions go through approval.

पाठ: Case Study: A Refund Assistant

26. Which item belongs on a pre-publish checklist?

  1. Every loop has an iteration cap and the cap path is handled
  2. Give every agent every tool
  3. Leave else branches empty
  4. Skip evaluation for small changes
जवाब देखें

जवाब: Every loop has an iteration cap and the cap path is handled

Bounded, tested, reversible.

पाठ: An Agent Workflow Checklist

← कोर्स पर वापस जाएँ