What should you focus on for an AI Engineer interview?

What makes this role different?

Agents, retrieval and system design stand out. These topics appeared more often in AI Engineer interviews than across all interviews. LLM fundamentals, evaluation, embeddings and vector databases also appeared more often. Most questions were mid-level or senior-level.

Can you explain how you would build an AI application, not just define its parts? Use a chatbot or enterprise assistant as your check. Can you connect model choice, retrieval, tool use, testing and deployment in the same design?

Can you explain a RAG pipeline from query to answer?

“How would you build a RAG pipeline?” was the most asked question, appearing in 19 of 190 interviews. Questions also covered what RAG is, how to choose chunk sizes and how to evaluate the pipeline before deployment.

In 6 of 10 reviewed answers, the explanation did not show how a query leads to retrieved passages, prompt assembly and a generated answer. Check that your answer follows that path.

  • Can you describe how documents are ingested, split into chunks and used to create embeddings?
  • Can you explain how the query leads to retrieved passages?
  • Can you show how those passages enter the prompt before the model generates an answer?
  • Can you explain how semantic search differs from keyword search?

Can you explain when an agent is needed?

The questions ask what agents are, whether a task needs one and when to choose LangGraph over LangChain.

Can you explain that choice without only naming a framework? Check whether you can describe how the agent calls tools, manages memory and context, and coordinates with other agents. MCP servers are another topic to check.

Can you explain how you would test and run the system?

Evaluation and production topics appeared in 85 of 190 interviews. Your checks should cover both the answers the application gives and how it runs.

  • How would you evaluate a RAG pipeline before deployment?
  • How would you monitor and troubleshoot the application after deployment?
  • How would you address latency, scaling and cost?
  • How would you handle prompt injection, privacy and human approval?

Can you write the code as well as explain it?

Coding tasks appeared in 30 of 190 interviews. Can you write a Two Sum solution, generate the Fibonacci series, demonstrate dynamic programming and write SQL to remove duplicate records?

Beyond those tasks, can you explain concurrency and asynchronous programming in Python? For backend work, can you describe how you would use retries to support reliable processing?

What AI Engineer interviews cover

System design & architecture59% of interviews
AI agents, tool calling & MCP51% of interviews
RAG & retrieval48% of interviews
LLM fundamentals35% of interviews
LLM evaluation, guardrails & hallucinations33% of interviews
Performance optimization & profiling24% of interviews

The 20 most asked AI Engineer questions

Counted by interview, not by mention: a question asked twice in one conversation counts once.

  1. How would you build a RAG pipeline?

    Asked in 19 of 190 interviews · last asked Oct 21 · Design · Mid-level

    Show answer

    I would build RAG by combining retrieval from external sources with an LLM that generates answers. First, I would clean documents, split them into chunks, and index their embeddings. For each query, I would retrieve relevant chunks, add them and the question to a prompt, and generate a cited answer. This helps with private or changing information without retraining.

    • Use vector search, keyword search, or both to retrieve relevant passages; rerank if needed.
    • Keep documents current and enforce access permissions before passing retrieved text to the model.
    • Ask for answers supported by sources, and say when the sources are insufficient.
    • Evaluate retrieval quality, answer accuracy, latency, and cost separately.

    What the interviewer checks: The interviewer checks whether you can explain the full path from documents and a user query to a grounded answer.

    Common mistake (in 6 of 10 reviewed answers): Many answers forget that retrieved passages must be added to the prompt so the LLM can use them.

    Other ways it was asked
    • “What is retrieval-augmented generation (RAG), and when would you use it?”
    • “What is retrieval-augmented generation (RAG)?”
    • “Can you explain the architecture of a retrieval-augmented generation (RAG) system?”

  2. What is RAG, and how is it used in an LLM system?

    Asked in 14 of 190 interviews · last asked Oct 23 · Concept · Junior

    Show answer

    RAG means retrieval-augmented generation. It retrieves relevant information from an external source and adds it to the prompt so an LLM can use it to answer. I use it for answers that need current or private information without retraining the model. It can reduce wrong answers, but it does not prevent them.

    • Retrieve relevant documents or passages, then include them with the user's question in the prompt.
    • Retrieval can use keyword search, vector search, or both.
    • Answer quality depends on retrieving useful information and the model using it correctly.
    • RAG does not change the model's weights.

    What the interviewer checks: The interviewer is checking whether you understand the retrieval-to-generation flow, when to use it, and its limits.

    Other ways it was asked
    • “What is RAG?”
    • “How does RAG work?”
    • “What is RAG, and what does it do?”

  3. How would you choose a chunking strategy and chunk size for a RAG system?

    Asked in 10 of 190 interviews · last asked Oct 21 · Scenario · Mid-level

    Show answer

    I would split documents at natural boundaries, such as headings or paragraphs, while keeping related information together. I would test chunk sizes against real questions, within the embedding model’s input limit and the generation model’s context budget. Smaller chunks give focused matches; larger chunks preserve context but may add noise. I would add overlap only if testing shows it helps.

    • Match chunk boundaries to document structure and the questions users ask.
    • Measure chunk size in tokens, not characters, against model limits.
    • Compare retrieval quality and answer accuracy across sizes; there is no single best size.
    • Overlap can preserve context across boundaries, but increases storage and duplicate results.

    What the interviewer checks: The interviewer is checking whether you balance context, retrieval quality, model limits, and cost through testing.

    Other ways it was asked
    • “How do you choose a document-chunking strategy for a RAG system?”
    • “What factors should you consider when designing chunking strategies for a RAG system?”
    • “How do you choose a chunking strategy and chunk size for a RAG system?”

  4. What is a large language model?

    Asked in 7 of 190 interviews · last asked Oct 1 · Concept · Junior

    Show answer

    A large language model (LLM) is a neural network trained on large amounts of text to learn language patterns. Most use transformers and learn by predicting next or missing tokens, which are pieces of text. Generative LLMs write text one token at a time, but their answers can be wrong.

    • Encoder-only models build text representations, often used for classification or search.
    • Decoder-only models predict the next token and are common in chat assistants.
    • Encoder-decoder models read input and generate output, often for translation or summarization.
    • Training adjusts model weights; generating an answer normally uses those weights without changing them.

    What the interviewer checks: The interviewer checks whether you understand what LLMs learn, how they generate text, and their main limitations.

    Other ways it was asked
    • “What is an LLM?”
    • “What are LLMs?”
    • “What are the main types of large language models?”

  5. How would you evaluate a RAG pipeline before deploying it?

    Asked in 7 of 190 interviews · last asked Oct 19 · Concept · Mid-level

    Show answer

    I would test the pipeline on realistic customer questions with checked answers and supporting documents. I would measure retrieval quality, answer correctness, and whether claims are supported by sources. I would also test missing evidence, access controls, latency, and cost under expected traffic. I would deploy only after it meets agreed targets.

    • Check whether retrieval finds the needed evidence, separately from judging the final answer.
    • Use customer-specific tests, including hard questions and questions the documents cannot answer.
    • Combine automated scoring with human review; check citations and unsupported claims.
    • Test access controls, prompt injection, response times under load, and cost per request.

    What the interviewer checks: The interviewer checks whether you can test retrieval, answer quality, safety, and production readiness separately.

    Other ways it was asked
    • “How would you implement a RAG system step by step?”
    • “How would you evaluate a RAG pipeline for each customer?”
    • “Can you explain a RAG system end to end?”

  6. How would you evaluate an LLM application?

    Asked in 6 of 190 interviews · last asked Oct 6 · Scenario · Mid-level

    Show answer

    I would test the whole application against clear task goals, not just the model. I would use realistic cases to measure answer quality, safety, response time, and cost. I would combine automated checks with human review, test failures and attacks, and monitor real user outcomes after launch.

    • Build a fixed test set with typical requests, difficult cases, and expected results or scoring rules.
    • Check correctness, relevance, and whether answers are supported by sources when sources are required.
    • Use human review to check model-based scoring; test prompt injection, privacy leaks, and harmful outputs.
    • Compare changes on the same tests, then monitor quality, response time, cost, and failures in production.

    What the interviewer checks: The interviewer wants to see whether you can measure user value and production risks, not just model accuracy.

    Other ways it was asked
    • “How would you evaluate an LLM system?”
    • “What parameters or techniques would you use to evaluate an LLM application?”
    • “How would you design and implement an LLM application?”

  7. What is semantic search, and how does it differ from keyword search?

    Asked in 5 of 190 interviews · last asked Oct 9 · Concept · Junior

    Show answer

    Semantic search finds results by meaning, not just shared words. It often uses embeddings to match related ideas, even when wording differs. Keyword search matches words or terms and is often better for exact names, codes, or phrases.

    • Semantic search can find relevant results that use different words from the query.
    • Similarity search finds nearby vectors; it supports semantic search when those vectors capture meaning.
    • Hybrid search combines keyword and semantic search to cover both exact matches and related meanings.
    • Check the model and retrieval method; results alone do not prove that a system uses semantic search.

    What the interviewer checks: The interviewer checks whether you understand meaning-based retrieval, word matching, and how vector similarity supports semantic search.

    Other ways it was asked
    • “What do you mean by semantic search?”
    • “What do you mean by semantic search, and how is it different from normal search?”
    • “What is the difference between semantic search and similarity search, and when do you use each?”

  8. When would you choose LangGraph over LangChain?

    Asked in 5 of 190 interviews · last asked Sep 27 · Concept · Senior

    Show answer

    I would choose LangGraph when I need explicit control over a stateful workflow, such as branching, loops, or human approval steps. I would use LangChain for model and tool integrations or a higher-level agent setup. LangChain agents run on LangGraph, so the two can work together.

    • LangGraph gives direct control over workflow steps, transitions, and shared state.
    • Configured checkpointing supports pausing and resuming work, including human approval.
    • LangChain offers higher-level abstractions; use LangGraph directly when you need more control.

    What the interviewer checks: The interviewer checks whether you can choose the right level of workflow control without treating the tools as competitors.

    Other ways it was asked
    • “What is the difference between LangChain and LangGraph?”
    • “When would you use LangChain, and when would you use LangGraph?”
    • “When would you choose LangGraph over LangChain for an LLM application?”

  9. What are AI agents, and how do you decide whether a task needs one?

    Asked in 5 of 190 interviews · last asked Oct 21 · Concept · Mid-level

    Show answer

    An AI agent chooses actions toward a goal and uses feedback to decide what to do next. In LLM applications, it often calls tools in a loop. I use an agent when the task needs decisions that depend on intermediate results. If the steps are predictable, I prefer a fixed workflow because it is easier to test and control.

    • Agents choose next steps; fixed workflows follow predefined paths.
    • Use agents when changing conditions make a fixed sequence too limiting.
    • Weigh flexibility against extra cost, delay, and risk of errors.
    • Limit tool permissions, set stopping rules, and require approval for high-impact actions.

    What the interviewer checks: The interviewer checks whether you understand agent autonomy and can choose it over simpler approaches for clear reasons.

    Other ways it was asked
    • “What is an AI agent and how does it work?”
    • “When should you use an AI agent?”
    • “What is an AI agent, and how should you think about its role in an AI application?”

  10. What makes a good prompt, and what information should you include when prompting an AI agent?

    Asked in 4 of 190 interviews · last asked Oct 21 · Concept · Mid-level

    Show answer

    A good prompt gives a clear task, useful context, and a clear description of a successful result. Include the input, constraints, output format, and examples when needed. For an AI agent, also state which tools and actions are allowed, when to ask for approval, and when to stop.

    • Give relevant context and clearly separate instructions from input data.
    • Define success, limits, and the required output format.
    • Set tool permissions and require approval for sensitive or irreversible actions.
    • Tell the agent how to handle missing information; test and improve the prompt using real tasks.

    What the interviewer checks: The interviewer is checking whether you can give clear instructions and set safe boundaries for an agent.

    Other ways it was asked
    • “For prompt engineering, what makes a good prompt?”
    • “What is prompt engineering, and when would you use it?”
    • “What makes a good AI prompt?”

  11. What is a vector database?

    Asked in 4 of 190 interviews · last asked Oct 9 · Concept · Junior

    Show answer

    A vector database stores embeddings, which are lists of numbers that represent data such as text or images. It indexes these vectors to find similar items quickly. A vector store is a broader term for storage with vector search; a vector database usually adds features such as persistence, filtering, and updates. People sometimes use these terms interchangeably.

    • Common uses include semantic search, recommendations, and retrieving context for RAG.
    • Approximate nearest neighbor search trades some accuracy for speed.
    • Options include Pinecone, Milvus, Qdrant, Weaviate, and PostgreSQL with the pgvector extension.
    • An embedding model creates vectors; the database stores and searches them.

    What the interviewer checks: The interviewer checks whether you understand embeddings, similarity search, and the role of vector storage in AI systems.

    Other ways it was asked
    • “What vector databases are currently available?”
    • “What are vector databases and vector stores, and how do they differ?”

  12. How do you use the Model Context Protocol (MCP), and what tasks does it help you perform?

    Asked in 4 of 190 interviews · last asked Sep 4 · Tooling · Mid-level

    Show answer

    MCP is a standard protocol that connects AI applications to external tools and data. I use an MCP client in the application to connect to servers that expose tools, resources, and prompt templates. It helps an assistant search documents, query databases, or run workflows. I limit access and require approval for sensitive actions.

    • MCP provides a shared interface, reducing custom integration work across AI applications.
    • Servers expose tools to call, resources to read, and reusable prompt templates.
    • The application controls tool execution; the model does not gain unrestricted access.
    • When building servers, define clear input schemas, validate requests, enforce permissions, and return useful errors.

    What the interviewer checks: The interviewer checks whether you understand MCP’s purpose, client-server design, practical uses, and access controls.

    Other ways it was asked
    • “What do you know about MCP, the Model Context Protocol, for tool use?”
    • “What is the Model Context Protocol (MCP), and what problem does it solve?”
    • “What is the Model Context Protocol (MCP), and what are its use cases?”

  13. What is the difference between agentic AI and AI agents?

    Asked in 4 of 190 interviews · last asked Sep 24 · Concept · Mid-level

    Show answer

    Agentic AI describes an approach where AI plans and takes actions toward a goal. An AI agent is a specific system that observes information, chooses actions, and carries them out. Agentic AI can use one agent or several; the terms overlap and are not used consistently.

    • Agentic AI describes goal-directed behavior; an AI agent is a system that performs it.
    • Multiple agents are optional, not a defining requirement.
    • Generative AI creates content; agentic systems may use it to plan and choose actions.
    • Agents can have limited autonomy and require human approval for important actions.

    What the interviewer checks: The interviewer is checking whether you distinguish an approach from a system without claiming a strict, universal boundary.

    Other ways it was asked
    • “What is agentic AI?”
    • “What is the difference between agentic AI and generative AI?”
    • “What is the basic difference between an AI agent and agentic AI?”

  14. What is an MCP server?

    Asked in 4 of 190 interviews · last asked Oct 9 · Concept · Junior

    Show answer

    An MCP server is a program that exposes tools, data, and prompt templates through the Model Context Protocol (MCP). An LLM application connects to it using an MCP client. This lets the application discover and use features such as file access or database queries through a shared protocol.

    • MCP stands for Model Context Protocol.
    • Servers can expose tools to run, resources to read, and prompt templates to reuse.
    • The application manages connections and permissions; the model does not directly connect to the server.
    • Servers can run locally or remotely.

    What the interviewer checks: The interviewer checks whether you understand how MCP connects LLM applications to external tools and data.

    Other ways it was asked
    • “What does an MCP server do?”
    • “What are MCP servers, and how are they used in LLM applications?”
    • “What does MCP stand for?”

  15. How would you evaluate an AI agent, and which metrics would you use?

    Asked in 4 of 190 interviews · last asked Oct 16 · Concept · Senior

    Show answer

    I would test the agent on realistic tasks with clear success rules, including failure cases. I would measure task success, tool-use correctness, safety violations, latency, and cost per task. I would repeat tests to check reliability and review action logs to find failures. Human review helps judge results that automated checks cannot verify.

    • Measure completed tasks, not just how convincing the final answer sounds.
    • Check tool choices, arguments, and resulting actions, including unwanted changes.
    • Use repeated runs because the same task can produce different outcomes.
    • Compare versions on the same test set, with the same tools and budgets.

    What the interviewer checks: The interviewer is checking whether you can evaluate outcomes, agent actions, reliability, and practical trade-offs.

    Other ways it was asked
    • “How do you evaluate an AI agent?”
    • “Which metrics would you use to evaluate an AI agent?”
    • “How do you usually evaluate AI agents or agentic systems?”

  16. What is the difference between a list and a tuple in Python?

    Asked in 3 of 190 interviews · last asked Sep 9 · Concept · Junior

    Show answer

    A list is mutable: you can add, remove, or replace its items. A tuple is immutable: you cannot change its items after creation. I use lists for collections that may change and tuples for fixed groups of values, such as coordinates.

    • Lists use square brackets; tuples usually use parentheses, with commas defining the tuple.
    • Both preserve order and support indexing, slicing, and duplicate values.
    • A tuple can contain mutable objects, such as lists, whose contents can still change.
    • A tuple can be a dictionary key only if all its items are hashable; lists cannot.

    What the interviewer checks: The interviewer is checking whether you understand mutability and can choose the right collection.

    Other ways it was asked
    • “What is the difference between a Python list and a tuple?”
    • “What is the difference between a list and a tuple, and when would you use each?”

  17. How do multithreading and multiprocessing differ, and when would you choose each?

    Asked in 3 of 190 interviews · last asked Oct 21 · Concept · Junior

    Show answer

    Threads share memory within one process; processes have separate memory. In GIL-enabled CPython, only one thread runs Python code at a time. I choose threads for tasks that wait on network or disk, and processes for CPU-heavy Python work. Processes allow parallel execution but cost more memory and make sharing data harder.

    • Shared data needs careful access control with threads to avoid race conditions.
    • Processes add startup and communication costs; passing data often requires serialization.
    • Free-threaded CPython builds and native libraries that release the GIL can allow CPU parallelism with threads.

    What the interviewer checks: The interviewer checks whether you can choose based on workload, memory costs, and Python runtime limits.

    Other ways it was asked
    • “When would you choose multiple processes instead of multiple threads?”
    • “What is the difference between multiprocessing and multithreading in Python?”
    • “What is the difference between multithreading and multiprocessing?”

  18. What is the difference between == and is in Python?

    Asked in 3 of 190 interviews · last asked Sep 8 · Concept · Junior

    Show answer

    In Python, == compares values, while is checks whether two references point to the same object. Two separate objects can have equal values, so == can be True while is is False. Use is None to check for None. A single = assigns a value; it does not compare.

    • Use == for value equality and is for object identity.
    • Classes can customize == through __eq__; they cannot customize is.
    • Do not use is to compare numbers or strings; object reuse can make results misleading.
    Python
    a = [1, 2]
b = [1, 2]
c = a
print(a == b)  # True: equal values
print(a is b)  # False: separate objects
print(a is c)  # True: same object

    What the interviewer checks: The interviewer is checking whether you understand value equality, object identity, and assignment.

    Other ways it was asked
    • “In Python, what is the difference between = and ==?”
    • “What's the difference between == and is in Python?”

  19. What is the Global Interpreter Lock (GIL) in Python, and how does it affect concurrency?

    Asked in 3 of 190 interviews · last asked Aug 13 · Concept · Mid-level

    Show answer

    In GIL-enabled CPython, the Global Interpreter Lock lets only one thread execute Python bytecode at a time. Threads can overlap I/O waits, but they generally cannot run CPU-heavy Python code in parallel across cores. Separate processes can use multiple cores because each has its own interpreter and GIL.

    • The GIL is an implementation detail, not a rule for every Python implementation.
    • Some native libraries release the GIL, allowing CPU work to run in parallel.
    • CPython 3.13 and later offer optional free-threaded builds that can run without the GIL.
    • The GIL does not make shared data automatically safe; thread synchronization may still be needed.

    What the interviewer checks: The interviewer checks whether you distinguish concurrency from parallelism and choose threads or processes for the workload.

    Other ways it was asked
    • “How does Python's Global Interpreter Lock (GIL) work?”
    • “How does the Global Interpreter Lock affect parallelism in Python?”

  20. What is overfitting, how do you detect it, and how do you prevent it?

    Asked in 3 of 190 interviews · last asked Oct 23 · Concept · Mid-level

    Show answer

    Overfitting happens when a model learns training details or noise that do not generalize to new data. I detect it when training performance improves but validation performance stalls or gets worse. I reduce it with simpler models, regularization, early stopping, or more varied training data.

    • Use validation data that reflects real use, and prevent data leakage.
    • Learning curves show the gap between training and validation performance.
    • Tune on validation data; reserve the test set for final evaluation.
    • Underfitting means poor performance on both sets; consider better features, a more flexible model, or longer training.

    What the interviewer checks: The interviewer checks whether you can spot poor generalization and choose ways to improve it without misusing test data.

    Other ways it was asked
    • “How do you detect overfitting, and how do you prevent it?”
    • “What is overfitting, and how can you prevent it?”
    • “How can we overcome overfitting and underfitting?”

Python and coding interview questions

Python questions cover language basics, data structures, algorithms, coding tasks, concurrency and asynchronous programming. 116 questions.

  1. How would you write a function to solve the Two Sum problem?

    Asked in 2 of 190 interviews · last asked Oct 21 · Write code · Junior

    Other ways it was asked
    • “Write a function that solves the two-sum problem.”
    • “Write code for the Two Sum problem.”
    • “How would you approach solving the Two Sum problem and running your solution?”

  2. Can you tell me the difference between a shallow copy and a deep copy in Python?

    Asked in 2 of 190 interviews · last asked Aug 16 · Concept · Mid-level

    Other ways it was asked
    • “Can you explain deep copy and shallow copy in Python?”
    • “In Python, what's the difference between a shallow copy and a deep copy of a nested list or dict, and when can a shallow copy cause bugs?”

  3. What is the difference between mutable and immutable objects?

    Asked in 2 of 190 interviews · last asked Aug 22 · Concept · Junior

    Other ways it was asked
    • “What is the difference between mutable and immutable data types, and where would you use them?”

  4. Write Python code to generate the Fibonacci series.

    Asked in 2 of 190 interviews · last asked Aug 17 · Write code · Junior

    Other ways it was asked
    • “Write a Python program to display the Fibonacci series.”
    • “Write a Python program for the Fibonacci series.”
    • “Can you write Python code to generate the Fibonacci sequence?”

  5. How do you define a list in Python?

    Asked in 2 of 190 interviews · last asked Sep 8 · Concept · Junior

    Other ways it was asked
    • “What is a list in Python?”
    • “Can you explain what a list is in Python?”

  6. What is recursion?

    Asked in 2 of 190 interviews · last asked Sep 8 · Concept · Junior

    Other ways it was asked
    • “Can you explain recursion?”

  7. Can you write a Python program to demonstrate dynamic programming?

    Asked in 2 of 190 interviews · last asked Jul 17 · Write code · Mid-level

    Other ways it was asked
    • “Give an example code that demonstrates dynamic programming in Python.”

  8. What is a lambda function in Python, and how do you use it?

    Asked once · last asked Sep 8 · Concept · Junior

    Other ways it was asked
    • “What is a lambda function?”

  9. Given an array containing only zeros and ones, how would you find the maximum distance between two ones?

    Asked once · last asked Aug 22 · Write code · Junior

  10. How does dependency injection work in Python, and when would you use it?

    Asked once · last asked Aug 13 · Concept · Mid-level

    Other ways it was asked
    • “Why is dependency injection used in Python applications?”

  11. What is a common pitfall when using mutable default arguments in Python?

    Asked once · last asked Aug 16 · Concept · Mid-level

    Other ways it was asked
    • “How do Python's mutable default arguments behave in function definitions, and how do you avoid the common pitfall?”

  12. What is the binary search algorithm, and when can it be used?

    Asked once · last asked Aug 20 · Concept · Junior

    Other ways it was asked
    • “What is binary search, and when can it be used?”
Show 4 more

  1. How would you solve the longest palindromic subsequence problem using dynamic programming?

    Asked once · last asked Aug 22 · Write code · Mid-level

  2. How do you choose among multiprocessing, multithreading, and asynchronous programming for concurrent workloads in Python?

    Asked once · last asked May 13 · Concept · Mid-level

    Other ways it was asked
    • “How do you choose among concurrency, multiprocessing, multithreading, and asynchronous programming in Python?”

  3. Is there a difference between applying Python multithreading to I/O-bound versus CPU-bound workloads?

    Asked once · last asked May 13 · Concept · Mid-level

  4. Write a Python function that takes a PDF filename, extracts its contents, and summarizes them using an OpenAI model.

    Asked once · last asked Jun 12 · Write code · Mid-level

All Python interview questions

Machine learning and LLM interview questions

These questions cover AI concepts, learning algorithms, model training, transformers, tokens, prompting, fine-tuning and choosing a model. 205 questions.

  1. What is the basic difference between RAG and fine-tuning, and when would you use fine-tuning or RAG?

    Asked in 3 of 190 interviews · last asked Oct 21 · Concept · Mid-level

    Other ways it was asked
    • “When would you choose RAG over fine-tuning?”
    • “Can you briefly explain the difference between RAG and fine-tuning?”

  2. Why do LLMs hallucinate?

    Asked in 3 of 190 interviews · last asked Aug 19 · Concept · Mid-level

    Other ways it was asked
    • “Why do LLMs hallucinate, and what causes these hallucinations?”
    • “Why do LLMs hallucinate sometimes?”
    • “What do you believe triggers hallucinations?”

  3. What is supervised learning?

    Asked in 3 of 190 interviews · last asked Sep 24 · Concept · Junior

    Other ways it was asked
    • “What are supervised and unsupervised learning?”
    • “Explain supervised learning.”

  4. What is the difference between one-shot, few-shot, and zero-shot prompting?

    Asked in 2 of 190 interviews · last asked Aug 19 · Concept · Mid-level

    Other ways it was asked
    • “What are zero-shot, one-shot, and few-shot prompting, and when would you use each?”

  5. What is the difference between F1 and ROC-AUC?

    Asked in 2 of 190 interviews · last asked Aug 12 · Concept · Mid-level

    Other ways it was asked
    • “What are the ROC curve and the AUC metric, and how do they differ?”

  6. What is a KV cache, and what does it store during transformer inference?

    Asked in 2 of 190 interviews · last asked Aug 20 · Concept · Mid-level

    Other ways it was asked
    • “What is KV caching, and how does it work in LLM inference?”

  7. What are overfitting and underfitting?

    Asked in 2 of 190 interviews · last asked Aug 22 · Concept · Junior

    Other ways it was asked
    • “What is the difference between overfitting and underfitting?”

  8. What is the difference between LLM and AI?

    Asked in 2 of 190 interviews · last asked Jul 25 · Concept · Junior

    Other ways it was asked
    • “What is the difference between an AI agent and an LLM?”

  9. What is the difference between machine learning and deep learning?

    Asked in 2 of 190 interviews · last asked Aug 6 · Concept · Junior

  10. Can AI systems be built using open-source or closed-source models, and how do those approaches differ?

    Asked in 2 of 190 interviews · last asked Oct 13 · Concept · Junior

    Other ways it was asked
    • “What is the difference between open-source and closed-source AI models?”

  11. Can you explain what LLM and RAG are, and what two types of tokens are available in agentic AI?

    Asked in 2 of 190 interviews · last asked Jul 17 · Concept · Junior

    Other ways it was asked
    • “Can you explain LLMs, RAG, and tokens?”

  12. What is the difference between bagging and boosting?

    Asked once · last asked Jul 20 · Concept · Junior

    Other ways it was asked
    • “Could you describe the difference between boosting and bagging models?”
Show 5 more

  1. What is model fine-tuning, and when would you use it?

    Asked once · last asked Jul 29 · Concept · Mid-level

  2. How do you handle an imbalanced dataset?

    Asked once · last asked Jul 20 · Scenario · Mid-level

  3. How would you start building an ML pipeline to classify reviews and call transcripts by sentiment, and what approach would you take first?

    Asked once · last asked Jul 27 · Design · Mid-level

    Other ways it was asked
    • “How would you start building an ML pipeline to classify reviews and call transcripts?”

  4. What is distributed cognition, and how does it differ from extended cognition?

    Asked once · last asked Aug 12 · Concept · Mid-level

    Other ways it was asked
    • “How would you briefly summarize the difference between extended cognition and distributed cognition?”

  5. How would you solve the N-Queens problem using reinforcement learning?

    Asked once · last asked Aug 13 · Scenario · Mid-level

Backend, SQL and data engineering interview questions

Backend topics include databases, SQL, data pipelines, web applications, APIs, integrations, retries and reliable processing. 183 questions.

  1. What is the difference between the HTTP PUT and PATCH methods?

    Asked in 2 of 190 interviews · last asked Oct 13 · Concept · Junior

    Other ways it was asked
    • “What is the difference between the HTTP POST and PUT methods?”
    • “What is the difference between PUT and PATCH in a REST API?”

  2. What is the difference between batch processing and stream processing?

    Asked in 2 of 190 interviews · last asked Oct 19 · Concept · Mid-level

    Other ways it was asked
    • “What are stream processing and batch processing, and when would you use each?”
    • “What is the difference between batch processing and streaming?”

  3. How would you prevent duplicate SMS messages when processing repeated requests?

    Asked in 2 of 190 interviews · last asked Jul 27 · Concept · Senior

    Other ways it was asked
    • “How would you prevent duplicate SMS messages from being sent?”

  4. Why might an application route weather reminders using a temperature metric other than the daily minimum?

    Asked in 2 of 190 interviews · last asked Jul 27 · Concept · Mid-level

    Other ways it was asked
    • “Why might a weather alert system use a temperature metric other than the daily minimum?”

  5. What is FastAPI?

    Asked in 2 of 190 interviews · last asked Sep 21 · Concept · Junior

    Other ways it was asked
    • “What exactly is FastAPI?”
    • “What is the difference between Flask and FastAPI?”

  6. What is a data warehouse?

    Asked in 2 of 190 interviews · last asked Oct 19 · Concept · Junior

  7. How would you write SQL to remove duplicate records?

    Asked in 2 of 190 interviews · last asked Oct 19 · Write code · Junior

    Other ways it was asked
    • “How do you remove duplicate records from a database table using SQL?”

  8. What's the difference between WHERE and HAVING?

    Asked once · last asked Sep 8 · Concept · Junior

    Other ways it was asked
    • “What is the difference between WHERE and HAVING in SQL?”

  9. What is the N+1 query problem, and how can you avoid it?

    Asked once · last asked Aug 13 · Concept · Mid-level

    Other ways it was asked
    • “How would you avoid the N+1 query problem?”

  10. What are SQL window functions, and when would you use one?

    Asked once · last asked Aug 20 · Concept · Mid-level

    Other ways it was asked
    • “What are SQL window functions, and when would you use them?”

  11. What is the difference between ETL and ELT, and when would you use each in data pipeline development?

    Asked once · last asked Oct 19 · Concept · Mid-level

    Other ways it was asked
    • “What is the difference between ETL and ELT?”

  12. What are the disadvantages of database indexes?

    Asked once · last asked Aug 13 · Concept · Mid-level

Show 8 more

  1. What is the difference between OLTP and OLAP systems?

    Asked once · last asked Aug 20 · Concept · Junior

    Other ways it was asked
    • “What are OLAP and OLTP systems, and how do they differ?”

  2. How is OCR integration used in an application?

    Asked once · last asked Oct 23 · Tooling · Mid-level

    Other ways it was asked
    • “How would you use OCR in an application?”

  3. How would you integrate an enterprise financial system with Office 365?

    Asked once · last asked Oct 8 · Design · Senior

    Other ways it was asked
    • “How would you integrate an enterprise financial system with Office 365 so that actions in one system are correctly performed in the other?”
    • “What challenges can arise when integrating an enterprise financial system with Office 365 so that actions are correctly performed in both systems, and how would you address them?”

  4. Can you explain the first three normal forms in database normalization?

    Asked once · last asked Sep 8 · Concept · Mid-level

  5. How do you pass user-entered text from a React application to a backend API?

    Asked once · last asked Jul 31 · Scenario · Mid-level

    Other ways it was asked
    • “How do you pass user text from React state to a backend API?”

  6. How would you use SQL to find the latest record for each user?

    Asked once · last asked Aug 12 · Write code · Junior

    Other ways it was asked
    • “How would you find the latest record per user in SQL?”

  7. When importing database rows into an application, how would you generate a unique key and validate missing values?

    Asked once · last asked Aug 22 · Scenario · Mid-level

    Other ways it was asked
    • “When importing database rows into an application, how would you generate a unique key and assign values for missing fields?”

  8. How would you design rate limiting for a public API?

    Asked once · last asked Aug 22 · Design · Senior

RAG questions follow documents through ingestion, chunking, embeddings and retrieval, alongside vector databases, hybrid search and answers grounded in retrieved content. 164 questions.

  1. How does the Boolean retrieval model score document-query matches, and when does it return zero or one?

    Asked in 3 of 190 interviews · last asked Aug 3 · Concept · Junior

    Other ways it was asked
    • “When is similarity high or low in the Boolean information retrieval model?”
    • “When does the Boolean information retrieval model's similarity function return 1?”
    • “When is similarity high or low in the Boolean model?”

  2. What are embeddings?

    Asked in 3 of 190 interviews · last asked Oct 19 · Concept · Junior

    Other ways it was asked
    • “What is your understanding of embeddings?”
    • “What are embeddings, and where are they used?”

  3. How does the probabilistic retrieval model calculate document-query similarity, and what assumptions does it make?

    Asked in 2 of 190 interviews · last asked Jul 31 · Concept · Junior

    Other ways it was asked
    • “In the probabilistic information retrieval model, how is similarity between a document and a query calculated?”
    • “In the probabilistic information retrieval model, how do you calculate document-query similarity?”
    • “What is a similarity measure used to compare a query with a document in the probabilistic information retrieval model?”

  4. How do you choose an embedding model for a RAG system, what documents should be embedded, and why?

    Asked in 2 of 190 interviews · last asked Mar 16 · Concept · Senior

    Other ways it was asked
    • “Which embedding models do you know for RAG?”

  5. What is cosine similarity?

    Asked in 2 of 190 interviews · last asked Aug 19 · Concept · Junior

    Other ways it was asked
    • “What do you mean by cosine similarity?”

  6. What is chunking, and why is it required?

    Asked in 2 of 190 interviews · last asked Sep 21 · Concept · Junior

    Other ways it was asked
    • “Can you explain chunking?”

  7. How do you interpret how well a document matches a query, and how do you assess their similarity?

    Asked in 2 of 190 interviews · last asked Jul 31 · Concept · Junior

    Other ways it was asked
    • “In information retrieval, how do you interpret how well a document satisfies a query, and how do you evaluate document-query similarity?”

  8. What is the difference between agentic RAG and custom RAG, and when would you use each approach?

    Asked once · last asked Jul 23 · Concept · Senior

    Other ways it was asked
    • “How does custom RAG differ from agentic RAG?”
    • “What factors determine whether to use custom RAG or agentic RAG?”

  9. How does semantic search with FAISS differ from traditional search methods when handling large volumes of data?

    Asked once · last asked Jul 1 · Concept · Mid-level

    Other ways it was asked
    • “How can semantic search with FAISS improve retrieval accuracy or efficiency compared with traditional search methods?”

  10. What data would you store in the vector database for a ticket-handling agent?

    Asked once · last asked Aug 7 · Design · Mid-level

    Other ways it was asked
    • “What data would you store in a conventional database for a ticket-handling agent?”

  11. What are large language models (LLMs) and retrieval-augmented generation (RAG), and how do they differ?

    Asked once · last asked Aug 17 · Concept · Mid-level

    Other ways it was asked
    • “What are large language models (LLMs) and retrieval-augmented generation (RAG)?”

  12. How would you create a vector and use it for semantic search?

    Asked once · last asked Aug 19 · Write code · Mid-level

    Other ways it was asked
    • “How would you write code to create a vector and perform semantic search on it?”
Show 3 more

  1. How would semantic search find results for a word such as “good” in vector space?

    Asked once · last asked Aug 19 · Concept · Mid-level

    Other ways it was asked
    • “How does semantic search find results for a word such as “good” in vector space?”

  2. Can semantic search be performed without embeddings?

    Asked once · last asked Aug 19 · Concept · Mid-level

    Other ways it was asked
    • “What approach would you use to implement semantic search without embeddings?”

  3. Which library would you use to calculate cosine similarity and match items?

    Asked once · last asked Aug 19 · Tooling · Mid-level

All RAG interview questions

AI agents, LangGraph and MCP interview questions

Agent topics include LangGraph, tool calling, MCP servers, memory, context management and coordination between agents. 173 questions.

  1. What is the difference between a chatbot and an AI agent?

    Asked in 3 of 190 interviews · last asked Oct 6 · Concept · Junior

    Other ways it was asked
    • “What is the difference between a chatbot and an agent?”
    • “What is an AI agent, and how does it differ from a chatbot interface?”

  2. What distinguishes an AI agent from a fixed workflow?

    Asked in 2 of 190 interviews · last asked Aug 7 · Concept · Senior

    Other ways it was asked
    • “What distinguishes an AI agent from an autonomous workflow built from rigid rules?”
    • “Why might you choose an AI agent rather than a fixed workflow?”

  3. How would you design and manage the system prompt for an LLM agent?

    Asked in 2 of 190 interviews · last asked Jul 16 · Concept · Mid-level

    Other ways it was asked
    • “How would you prompt an LLM using a system prompt?”

  4. What are external tools in an AI agent, and how does MCP relate to them?

    Asked in 2 of 190 interviews · last asked Sep 8 · Concept · Mid-level

    Other ways it was asked
    • “What is MCP, and when would you use it to connect an AI agent to tools?”

  5. How would you build an MCP server, and what components would it include?

    Asked in 2 of 190 interviews · last asked Sep 8 · Design · Mid-level

    Other ways it was asked
    • “How would you architect an application using an MCP server, including the client-side components?”
    • “How would you design an MCP server for an application, and what three tools or capabilities would it expose?”

  6. When would you use a fixed workflow, a single agent, or a multi-agent system?

    Asked in 2 of 190 interviews · last asked Aug 13 · Concept · Senior

    Other ways it was asked
    • “Why might you build a fixed workflow when agent technology is available?”

  7. How would you provide an AI agent with relevant data sources and context for a task?

    Asked in 2 of 190 interviews · last asked Oct 21 · Scenario · Mid-level

    Other ways it was asked
    • “How would you provide an AI agent with the context it needs?”

  8. What are skills in AI agents, and how are they used?

    Asked in 2 of 190 interviews · last asked Oct 14 · Concept · Mid-level

    Other ways it was asked
    • “What is a skill in an AI agent?”
    • “How do you implement skills for an AI agent?”

  9. What information would you store in a LangGraph state object?

    Asked once · last asked Jul 20 · Design · Mid-level

    Other ways it was asked
    • “How does LangGraph pass information between nodes, and what does the state look like?”

  10. Write a prompt for a multilingual voice AI agent with agentic tool orchestration.

    Asked once · last asked Jul 27 · Mid-level

    Other ways it was asked
    • “How would you design and implement a multilingual voice AI agent with agentic tool orchestration?”

  11. How would you manage context in a long-running chat with a user?

    Asked once · last asked Jul 28 · Design · Senior

    Other ways it was asked
    • “How would you manage context in a long-running chat, including recent messages, summaries, and when to generate a summary?”

  12. What information would you store from past tickets in the memory of a ticket-handling agent?

    Asked once · last asked Aug 7 · Design · Senior

    Other ways it was asked
    • “What metadata would you store for past tickets in an agent's memory?”
Show 3 more

  1. How do you build an agent with frameworks like Google ADK or LangGraph?

    Asked once · last asked Aug 20 · Tooling · Mid-level

    Other ways it was asked
    • “How would you get started building an agent with Google ADK or LangGraph?”

  2. How would you decide which source to trust when tools return conflicting information?

    Asked once · last asked Sep 22 · Scenario · Senior

  3. When is MCP useful, and when would you use it?

    Asked once · last asked Feb 5 · Concept · Mid-level

All Agentic AI interview questions

AI system design interview questions

Design questions cover complete systems for chatbots, voice agents, multimodal applications, enterprise assistants and business automation. 97 questions.

  1. How would you design the high-level architecture of an AI agent?

    Asked in 2 of 190 interviews · last asked Jul 16 · Design · Mid-level

    Other ways it was asked
    • “How would you build an AI agent?”

  2. What steps would you take to take a generative AI project from scratch to production?

    Asked in 2 of 190 interviews · last asked Jul 14 · Design · Senior

    Other ways it was asked
    • “How would you take a generative AI application from initial design through production deployment?”

  3. A long-term care pharmacy software company has about 200 facilities using its software and handles around 400 support tickets a day, with 40% covering repeated questions. Its agents have trouble finding answers in the documentation. Can AI help, and how would you approach this?

    Asked in 2 of 190 interviews · last asked May 13 · Design · Junior

    Other ways it was asked
    • “How would you design an AI system to help a long-term-care pharmacy support team find answers in its documentation faster, serving about 200 facilities and handling 400 tickets a day, with roughly 40% repetitive tickets?”

  4. How would you design a maintainable, evaluable LLM system?

    Asked in 2 of 190 interviews · last asked Sep 9 · Design · Mid-level

    Other ways it was asked
    • “How would you build an evaluation framework for a production LLM system?”

  5. How would you design a production multi-agent system end to end?

    Asked in 2 of 190 interviews · last asked Oct 7 · Design · Senior

    Other ways it was asked
    • “How would you orchestrate a multi-agent system?”

  6. How would you design an AI system to determine whether a site unit is ready for site acceptance testing, identify blockers, and provide the right contact?

    Asked once · last asked Sep 29 · Design · Senior

    Other ways it was asked
    • “How would you design an AI system to assess a site unit's readiness for site acceptance testing?”
    • “How would you design a real-time AI system to determine whether a site unit is ready for site acceptance testing, identify blockers, and provide the right contact?”
    • “How would you build an AI system for a field engineer to check whether a site unit is ready for site acceptance testing?”

  7. How would you design an AI system to enhance customer support?

    Asked once · last asked Jun 22 · Design · Mid-level

    Other ways it was asked
    • “Which technique would you use to build an AI system that enhances customer support?”

  8. What information should an AI phone agent have to handle each insurance call effectively?

    Asked once · last asked Jul 16 · Design · Senior

  9. What should an AI phone agent do after dialing an insurer's 1-800 number and the call starts ringing?

    Asked once · last asked Jul 16 · Scenario · Mid-level

  10. How would you design an AI bot that converts natural-language questions into safe SQL queries against a customer database?

    Asked once · last asked Sep 25 · Design · Senior

    Other ways it was asked
    • “How would you design an AI bot that accepts natural-language questions, queries a customer's SQL database, and responds through a website?”

  11. If the application has direct database access, why use an LLM, and how would you design the solution?

    Asked once · last asked Sep 25 · Design · Senior

    Other ways it was asked
    • “If we have direct database access, why use an LLM to query the database, and why make the design complicated?”

  12. How do you approach scaling beyond small automations to larger, robust business systems?

    Asked once · last asked Sep 29 · Design · Senior

Show 8 more

  1. How would you build an assistant that provides accurate, up-to-date, cited, access-controlled answers from multiple data sources?

    Asked once · last asked Sep 29 · Design · Senior

    Other ways it was asked
    • “How would you build an assistant that provides accurate, up-to-date, cited, access-controlled answers across more than 100 tools and data sources?”

  2. How would you approach screen analytics to determine what an agent is doing and how much time they spend in applications such as Outlook or Facebook?

    Asked once · last asked Oct 7 · Design · Senior

    Other ways it was asked
    • “How would you use screen analytics to measure an agent's activity, time spent in applications, and metrics such as emails sent?”

  3. How would you handle an agent that uses both static company-policy information and dynamic patient or insurance data, and where would fine-tuning fit?

    Asked once · last asked Feb 5 · Design · Senior

  4. How would you integrate speech-to-text and text-to-speech with a PBX?

    Asked once · last asked Feb 19 · Scenario · Mid-level

  5. For speech-to-text and text-to-speech integrations with a PBX, when would you process audio in real time versus in batches?

    Asked once · last asked Feb 19 · Concept · Mid-level

  6. How would you design the architecture of an AI system?

    Asked once · last asked Mar 12 · Design · Senior

    Other ways it was asked
    • “Tell me about an AI system you built and explain its architecture.”

  7. What role can a RAG pipeline play in a voice-based AI platform, and why might you use one?

    Asked once · last asked Mar 12 · Design · Senior

  8. What is an example of an employee health benefits process that could be automated?

    Asked once · last asked Apr 7 · Case

LLM evaluation and MLOps interview questions

Evaluation and production questions cover model and agent testing, deployment, CI/CD, monitoring, troubleshooting, scaling, latency and cost. 206 questions.

  1. How would you design a CI/CD pipeline?

    Asked in 3 of 190 interviews · last asked Jul 30 · Design · Mid-level

    Other ways it was asked
    • “How does a CI/CD pipeline work?”
    • “How does a CI/CD pipeline work, and what are its main stages?”
    • “How would you design a framework for triggering CI/CD pipelines?”

  2. How would you measure whether fine-tuning improved the model?

    Asked in 3 of 190 interviews · last asked Sep 27 · Scenario · Mid-level

    Other ways it was asked
    • “How would you determine whether fine-tuning actually improves a model?”
    • “How would you systematically compare a model's performance before and after fine-tuning?”

  3. How would you reduce token usage in an LLM application?

    Asked in 3 of 190 interviews · last asked Sep 8 · Scenario · Mid-level

    Other ways it was asked
    • “What strategies do you use to reduce token usage in an LLM application?”
    • “How can you reduce LLM token costs without sacrificing quality?”

  4. How would you deploy a machine learning model, validate its quality, and monitor for degradation?

    Asked in 2 of 190 interviews · last asked Oct 19 · Scenario · Mid-level

    Other ways it was asked
    • “How would you take a machine learning model from development to production?”
    • “How do you take a new model from training to production?”
    • “How do you take a new model from training to production, including quality gates, drift detection, and detection of silent degradation?”

  5. How would you reduce latency in a voice-based AI platform?

    Asked in 2 of 190 interviews · last asked Sep 24 · Scenario · Senior

    Other ways it was asked
    • “How would you reduce latency in an AI system?”

  6. How would you trace an AI agent's actions, and when would tracing be useful?

    Asked in 2 of 190 interviews · last asked Oct 8 · Scenario · Senior

    Other ways it was asked
    • “How would you track AI agents and the actions they perform?”
    • “How would you track more than 700 AI agents and the actions they perform?”
    • “What is tracing for AI agents, and when would you use it?”

  7. What techniques can reduce LLM API costs, and how would you measure the savings?

    Asked in 2 of 190 interviews · last asked Oct 7 · Scenario · Senior

    Other ways it was asked
    • “What techniques can reduce latency and API costs in an LLM application?”

  8. How would you structure a test automation framework, and what changes are needed when transitioning it to AI-based testing?

    Asked once · last asked Jul 30 · Design · Senior

    Other ways it was asked
    • “How would you structure a test automation framework, including its folder structure, and what changes are needed when transitioning it to AI-based testing?”

  9. How would you debug a chatbot that is slow and sometimes gives incorrect answers about product specifications?

    Asked once · last asked Jul 16 · Scenario · Mid-level

  10. How would you implement a chatbot if the RAG pipeline takes a long time to respond?

    Asked once · last asked Sep 10 · Scenario · Senior

    Other ways it was asked
    • “How would you implement asynchronous processing for a chatbot with a slow RAG pipeline beyond simply using an async function or queue?”
    • “How should a chatbot determine when to trigger a RAG pipeline?”

  11. What optimizations would you make to reduce tool calls or scale the tool execution pipeline?

    Asked once · last asked Sep 22 · Design · Senior

    Other ways it was asked
    • “What optimizations would you implement to reduce tool call costs?”

  12. How do you monitor a RAG system in production over time, and what do you do if you find issues?

    Asked once · last asked Oct 14 · Scenario · Senior

    Other ways it was asked
    • “How do you monitor a RAG system in production, identify issues, and address them?”
Show 5 more

  1. How would you assess whether an LLM is performing well, and what metrics or parameters would you use to validate it?

    Asked once · last asked Feb 5 · Scenario · Mid-level

  2. If you were assigned a project, what measures would you take to control both cost and response time?

    Asked once · last asked Feb 5 · Scenario · Mid-level

  3. What design patterns would you use to scale an AI agent to 10,000 users?

    Asked once · last asked Apr 17 · Design · Senior

  4. How would you integrate or deploy a Granite model with Watson AI on AWS GovCloud, and what architecture would you choose among AWS-native services, inference containers, or APIs?

    Asked once · last asked Apr 20 · Design · Senior

    Other ways it was asked
    • “How did you integrate or deploy the Granite model with Watson AI on AWS GovCloud? What was the integration architecture, and did you use AWS-native services, inference containers, or APIs to run it?”

  5. How would you deploy an AI agent application?

    Asked once · last asked Apr 29 · Scenario · Mid-level

AI security and governance interview questions

Security topics include authentication, application security, prompt injection, privacy, guardrails, regulated workflows and human approval controls. 98 questions.

  1. What is prompt injection, and how can you protect a system against it?

    Asked in 2 of 190 interviews · last asked Aug 19 · Scenario · Senior

    Other ways it was asked
    • “What is prompt injection?”

  2. What does a timestamp check protect against when verifying a webhook?

    Asked in 2 of 190 interviews · last asked Jul 27 · Concept · Senior

    Other ways it was asked
    • “Why should a webhook handler check a request timestamp?”

  3. How does webhook signature verification work?

    Asked in 2 of 190 interviews · last asked Jul 27 · Concept · Mid-level

    Other ways it was asked
    • “Why should a webhook handler verify a request signature?”

  4. How would you test a login feature for functionality, security, and edge cases?

    Asked once · last asked Sep 4 · Scenario · Junior

    Other ways it was asked
    • “How would you approach testing a new login feature to ensure it is both secure and functional?”
    • “Which edge cases would you test first for a new login feature?”

  5. What is the difference between JWT authentication and session-based authentication?

    Asked once · last asked Aug 22 · Concept · Mid-level

    Other ways it was asked
    • “What is the difference between JWT, session-based authentication, and OAuth?”

  6. What measures would you implement to secure a cloud-hosted PBX?

    Asked once · last asked Feb 19 · Scenario · Mid-level

    Other ways it was asked
    • “When securing a cloud-hosted PBX, which security controls would you implement yourself and which would you rely on an existing setup to provide?”
    • “How would you use a firewall to secure a cloud-hosted PBX?”

  7. How would you explain the potential risks of AI to a politician concerned about AI regulation?

    Asked once · last asked Jul 29 · Scenario · Mid-level

    Other ways it was asked
    • “How would you explain the importance of AI safety risks to a politician without relying on technical details?”

  8. How would you verify an AI agent's SQL read and write actions before allowing them to reach production?

    Asked once · last asked Sep 8 · Scenario · Senior

    Other ways it was asked
    • “How would you verify an AI agent's SQL read and write actions before production, for example with dry-run transactions or approval gates?”

  9. What settings or deployment choices can prevent data processed by an LLM from being sent externally or made public?

    Asked once · last asked Feb 5 · Scenario · Mid-level

    Other ways it was asked
    • “What settings or deployment choices can prevent an LLM's data from being sent externally or made public?”

  10. How would you design human-in-the-loop governance for an agentic system, including where to draw the line between autonomous action and human involvement?

    Asked once · last asked Apr 17 · Scenario · Senior

    Other ways it was asked
    • “How would you think about governance, particularly human-in-the-loop governance, in an agentic system? Where would you draw the line between autonomous action and involving a human?”

  11. How would you protect users' personally identifiable information when sending data to an LLM?

    Asked once · last asked May 12 · Scenario · Mid-level

  12. How would you keep encrypted biometric images and temporary buffers secure and efficient in a Swift iOS application?

    Asked once · last asked Jul 1 · Scenario · Mid-level

Show 8 more

  1. When balancing privacy requirements with the performance benefits of cloud-based solutions, how do you decide which aspect should take precedence in a given project?

    Asked once · last asked Jul 1 · Scenario · Senior

  2. How do you store secrets and keep them secure in AI agents or applications?

    Asked once · last asked Jul 16 · Scenario · Mid-level

    Other ways it was asked
    • “How do you provide secrets to an AI agent, and how does it retrieve them at runtime?”
    • “How can an AI agent retrieve secrets securely without including them in its context?”

  3. What security and scalability best practices do you follow when designing an API for another system to call?

    Asked once · last asked Jul 16 · Design · Senior

  4. How would you develop processes and automations to improve different AI models while complying with NDAs?

    Asked once · last asked Jul 16 · Scenario · Mid-level

  5. What is an AI risk management framework?

    Asked once · last asked Jul 21 · Concept · Junior

  6. Could you explain the controls in the AI risk management framework?

    Asked once · last asked Jul 21 · Concept · Mid-level

  7. How would you verify the authenticity of an incoming webhook request?

    Asked once · last asked Jul 26 · Concept · Senior

  8. Why should you not disable webhook signature verification when a verification helper is unavailable?

    Asked once · last asked Jul 26 · Scenario · Senior

AI engineer behavioral and leadership interview questions

Behavioral questions cover project ownership, everyday AI tool use, stakeholder communication, team leadership, delivery trade-offs, adoption and business impact. 106 questions.

  1. How do you use AI in your work?

    Asked in 3 of 190 interviews · last asked Oct 8 · Tooling

    Other ways it was asked
    • “How would you use AI coding tools in software development?”
    • “How do you use AI in your work, and how do you assess the quality of its output?”
    • “What have you built, and how do you use AI in your work?”

  2. Tell me about a time an AI tool produced an unsafe or incorrect output. What did you do to resolve it?

    Asked once · last asked Oct 18 · Behavioral · Mid-level

  3. How do you use GitHub Copilot, and what benefits do you get from your preferred way of interacting with it?

    Asked once · last asked Aug 22 · Tooling · Mid-level

    Other ways it was asked
    • “How do you use GitHub Copilot in your daily development workflow?”

  4. Tell me about a work task where an AI tool produced useful output end to end. What input did you provide, and how did you validate and use the output?

    Asked once · last asked Oct 18 · Behavioral · Mid-level

  5. How would you assess whether an insurance company's employee health benefits are more competitive than those of other insurers?

    Asked once · last asked Apr 7 · Case · Mid-level

  6. How would you decide whether an employee-first health benefits workflow needs software automation?

    Asked once · last asked Apr 7 · Scenario · Mid-level

  7. How would you identify and prioritize the biggest automation gaps in a business?

    Asked once · last asked Apr 7 · Scenario · Mid-level

  8. How would you raise the average employee experience rating for health insurance claims from 7 to 9 or higher?

    Asked once · last asked Apr 9 · Case · Mid-level

  9. Given CSAT ratings and qualitative feedback about health insurance claims, what analysis would you conduct?

    Asked once · last asked Apr 9 · Case · Mid-level

  10. How would you use Codex, what would you automate, and which usage patterns would you consider most effective?

    Asked once · last asked Apr 17 · Tooling · Mid-level

    Other ways it was asked
    • “How did you use Codex, how much automation did you use, and what patterns did you find most effective?”

  11. A customer wants an AI agent for every workflow process, and a BI lead proposes replacing citizen developers’ Power BI and Tableau reports with a broader solution. How would you break the request into a scoped engagement?

    Asked once · last asked Apr 20 · Scenario · Senior

  12. What guardrails or checkpoints would you establish during an AI engagement to help ensure the solution moves beyond discovery or an MVP and is adopted?

    Asked once · last asked Apr 20 · Scenario · Senior

Show 8 more

  1. Tell me about a time you implemented an AI solution, from identifying the need through implementation. What challenges and stakeholder or requirement changes did you handle, what were the results, and what would you improve?

    Asked once · last asked Apr 27 · Behavioral · Senior

  2. How would you drive adoption of an AI workflow while preventing erroneous notifications from overwhelming users and causing alert fatigue?

    Asked once · last asked Apr 27 · Scenario · Senior

  3. How would you respond to an unhappy client, and can you describe a time you handled a similar situation?

    Asked once · last asked May 6 · Scenario

  4. How would you tell a client that an error was your fault if doing so could put the project at risk?

    Asked once · last asked May 6 · Scenario

  5. How would you manage and deliver a chatbot solution as a team, from pilot to production?

    Asked once · last asked May 13 · Behavioral · Senior

    Other ways it was asked
    • “Tell me about how you delivered a chatbot solution as a team from pilot to production.”

  6. How do you work with business analysts day to day, and how do you resolve delivery or requirements issues with them?

    Asked once · last asked May 13 · Behavioral · Mid-level

  7. What are your thoughts on progressing from proof-of-concept projects to production-scale projects?

    Asked once · last asked May 13 · Design · Senior

  8. How do Agile and Scrum practices structure team roles and cadence, and how does Kanban differ?

    Asked once · last asked May 13 · Concept · Mid-level

How this list is made

Every question here was asked by a real interviewer in a real job interview that a candidate ran MeetAssist in. After the call, the interviewer's questions are pulled from the transcript. Practice sessions, tests, exams and anything that was not a job interview are dropped.

Questions are rewritten so they make sense on their own: names of people, employers and clients are removed, and speech-to-text mistakes are fixed. Small talk, logistics and questions about one candidate's own résumé are left out. Rephrasings of the same question are merged, and the count is the number of different interviews it came up in.

The interviews come from candidates who use MeetAssist, so they lean towards software, data and AI roles: the shares describe these interviews, not every AI Engineer interview. Answers to the most asked questions are written and checked by an editor; a "common mistake" is shown only when reviews of real answers in these interviews show it, with the count.

190 interviews, 113 candidates, Nov 2025 to Oct 2026. Updated every month with new interviews.