Interview Practice Questions

What is the Singleton pattern and how do you prevent multiple instances from being created?

Explain the Factory pattern with a real-world example. When is it preferable to calling new directly?

How does the Decorator pattern differ from inheritance? Give a scenario where Decorator is superior.

What problem does the Observer pattern solve? Describe a real-world scenario where it's essential.

When would you use the Builder pattern instead of a traditional constructor? What problems does it solve?

Implement a thread-safe Singleton in Java using double-checked locking. What are the pitfalls, and what are better alternatives?

Design a payment processing system that can dynamically switch between PayPal, Stripe, and cryptocurrency providers. How would you avoid modification of the payment processor each time a new provider is added?

How does Spring's @Transactional annotation use the Proxy pattern under the hood? What happens when you call a @Transactional method from another method in the same class?

Implement an undo/redo system for a text editor using the Command pattern. How would you handle large undo stacks?

How would you design a notification system for a social media platform? Users should receive notifications for likes, comments, follows, and messages. Design for low latency and high throughput.

How does Java's InputStream chain (e.g., BufferedInputStream wrapping FileInputStream wrapping a file) use the Decorator pattern? Why is this design flexible?

Design a database connection pool. What patterns are involved? How do you prevent connection exhaustion and leaks?

Explain how Kafka's consumer group architecture uses the Observer pattern. How does it differ from traditional pub-sub?

How does the Composite pattern simplify working with file system trees? Give a concrete example of how you'd traverse and calculate total size.

Implement a logging framework that supports multiple outputs (console, file, remote syslog). How would you avoid code duplication and enable runtime configuration?

How would you implement a plugin architecture for a large application? Design for extensibility without modifying core code.

How does the State pattern model a TCP connection lifecycle (LISTEN, ESTABLISHED, CLOSE_WAIT, CLOSED)? Why is it better than a giant if/else statement?

Explain how the Chain of Responsibility pattern works in Express.js middleware. How do middleware functions pass control down the chain?

How does Retrieval-Augmented Generation (RAG) improve LLM factual accuracy? Design a RAG system for a customer support chatbot.

What is the ReAct (Reasoning + Acting) pattern? How does it combine reasoning with tool use to enhance LLM capabilities?

Design a rate limiter for an API gateway that can enforce per-user, per-endpoint, and global limits. How would you handle distributed rate limiting across multiple servers?

How would you implement a distributed cache with lazy loading? Design for cache misses, consistency, and handling of large datasets.

Design a CQRS (Command Query Responsibility Segregation) system for an e-commerce order service. Why separate writes and reads? How would you keep them synchronized?

Describe how Netflix's video player uses the State pattern to manage states like BUFFERING, PLAYING, PAUSED, ERROR. What are the state transitions?

Design an A/B testing framework for a large-scale web service. How would you track which variant a user sees, collect metrics, and ensure statistical validity?

Explain how the Template Method pattern structures the JUnit test lifecycle (setUp, test, tearDown). Why is this pattern ideal for test frameworks?

Design a multi-agent system for automated code review. How would multiple AI agents collaborate to identify bugs, suggest refactoring, and enforce style?

Design a feature store to prevent training/serving skew in a machine learning system. How would you ensure features are computed consistently offline and online?

Design a Mediator for a real-time collaborative document editor (like Google Docs). How would multiple users' edits be coordinated without direct object coupling?

Explain how the Visitor pattern enables adding operations to an Abstract Syntax Tree (AST) without modifying node classes. Why is this critical for compiler design?

You're building a RAG system over a 10-million-document corpus. Walk me through your chunking strategy, embedding approach, and retrieval design.

Name three ways a RAG pipeline can fail silently — where the system returns a confident-sounding answer that's wrong.

What is Chain of Thought prompting and why does adding 'Let's think step by step' improve LLM accuracy on reasoning tasks?

In production, Chain of Thought reasoning can increase cost and latency 3–5x. How would you get the accuracy benefits without paying the full price?

Describe the exact sequence of events in an LLM tool-use loop from the user's query to the final response. What can go wrong at each step?

An AI agent with tool access to send emails and update a database is deployed for customer support. What security controls would you put in place?

Design an Orchestrator-Worker system to generate a competitive analysis report for a given company. What workers do you define and how does the orchestrator coordinate them?

Compare the Orchestrator-Worker pattern to a peer-to-peer multi-agent mesh. When is each appropriate, and what are the failure mode differences?

Why would you implement token streaming in a chat application rather than waiting for the full LLM response? What are the implementation trade-offs?

What's the difference between input guardrails and output guardrails in an LLM system? Give two examples of each and explain when you'd apply them.

Describe the three types of memory in an AI agent: in-context, external (vector), and episodic. When does each type become insufficient?

Why is structured output critical for production AI systems? What are the three most common failure modes when an LLM returns free-form text instead of structured data?

You're building a multi-agent coding system where a Planner, Coder, and Reviewer agent collaborate. How do you handle the case where Reviewer rejects Coder's output and they get stuck in a loop?

Explain how prompt caching works in Anthropic's API and describe three production scenarios where it cuts cost by 50% or more.

A product manager asks why you can't just fine-tune the model to fix a recurring hallucination problem. Walk them through your decision framework for fine-tuning vs. better prompting vs. RAG.