Home/Questions/LLM

LLM Interview Questions 2026

Published July 22, 2026 · Updated July 22, 2026

6 LLM interview questions from Google, Microsoft, Meta, Anthropic. All difficulty levels, all roles. Upvoted by engineers who were asked them.

6 questions
Companies: Google, Microsoft, Meta, Anthropic
646 upvotes

LLM Questions by Company

Google (1)

Google Hard Scenario round 5–8 YearsLLMRAGPython

Our RAG pipeline responses are accurate in staging but hallucinate in production on long user queries. Walk me through how you debug and fix this.

↑ 112 upvotes · 89 engineers asked this · AI Engineer

Microsoft (1)

Microsoft Hard Scenario round 4–7 YearsLLMPython

You are building an AI agent that books travel. It has access to 6 tools. How do you prevent it from calling tools in the wrong order or entering infinite loops?

↑ 98 upvotes · 76 engineers asked this · AI Engineer

Meta (1)

Meta Hard Scenario round 5–8 YearsLLMPythonML/AI

Your fine-tuned LLM performs well on your eval set but users report degraded responses 2 weeks after deployment. What is happening and what do you do?

↑ 104 upvotes · 82 engineers asked this · AI Engineer

Anthropic (1)

Anthropic Hard Technical round 4–8 YearsLLMRAGPython

How would you reduce the latency of a RAG pipeline from 4 seconds to under 1 second in production?

↑ 119 upvotes · 94 engineers asked this · AI Engineer

OpenAI (1)

OpenAI Medium Technical round 2–5 YearsLLMPython

Explain the difference between zero-shot, few-shot, and chain-of-thought prompting. When would you choose each in production?

↑ 87 upvotes · 68 engineers asked this · AI Engineer

TikTok (1)

TikTok Hard Scenario round 6–10 YearsLLMML/AIPython

Design an AI content moderation system for a social platform with 10 million daily posts. How do you balance accuracy, cost, and latency?

↑ 126 upvotes · 97 engineers asked this · AI Engineer

Practice these questions with AI feedback

Get instant grading on your answers, identify your weak areas, and generate a personalised 14-day study plan — all free.

Build my study plan →

Frequently asked questions

Our RAG pipeline responses are accurate in staging but hallucinate in production on long user queries. Walk me through how you debug and fix this.
Our RAG pipeline responses are accurate in staging but hallucinate in production on long user queries. Walk me through how you debug and fix this.
You are building an AI agent that books travel. It has access to 6 tools. How do you prevent it from calling tools in the wrong order or entering infinite loops?
You are building an AI agent that books travel. It has access to 6 tools. How do you prevent it from calling tools in the wrong order or entering infinite loops?
Your fine-tuned LLM performs well on your eval set but users report degraded responses 2 weeks after deployment. What is happening and what do you do?
Your fine-tuned LLM performs well on your eval set but users report degraded responses 2 weeks after deployment. What is happening and what do you do?
How would you reduce the latency of a RAG pipeline from 4 seconds to under 1 second in production?
How would you reduce the latency of a RAG pipeline from 4 seconds to under 1 second in production?
Explain the difference between zero-shot, few-shot, and chain-of-thought prompting. When would you choose each in production?
Explain the difference between zero-shot, few-shot, and chain-of-thought prompting. When would you choose each in production?