ai-engineering-interview-questions

AI Engineering Interview Questions and Answers

AI Engineering Interview Questions and Answers - Your Cheat Sheet For AI Engineering Interviews

These interview questions and answers are helpful for roles such as:

• AI Engineer
• Gen AI Engineer
• LLM Engineer
• Agentic AI Engineer
• AI Agent Engineer
• Forward Deployed Engineer
• AI Solutions Architect
• AI Platform Engineer
• Applied AI Engineer
• MLOps Engineer
• LLMOps Engineer

Table of Contents

• Must Know
• LLM Fundamentals
• Prompt Engineering
• Retrieval-Augmented Generation (RAG)
• AI Agents and Agentic Systems
• Fine-Tuning and Model Adaptation
• Vector Databases and Embeddings
• AI System Design
• LLMOps and Production AI
• Evaluation and Testing
• AI Safety, Ethics, and Responsible AI
• Multimodal AI
• AI Infrastructure and Scalability
• Coding and Practical Implementation
• Behavioral and Scenario-Based Questions

Prepared and maintained by the Founder of Outcome School: Amit Shekhar

Follow Amit Shekhar

• X/Twitter
• LinkedIn
• GitHub

Follow Outcome School

• YouTube
• X/Twitter
• LinkedIn
• GitHub

I teach at Outcome School

• AI and Machine Learning

---

Note: We will keep updating this with new questions and answers.

---

Must Know

• LLM
• RAG
• MCP
• Agent
• Fine-tuning
• Quantization

Learn about the LLM, RAG, MCP, Agent, Fine-tuning & Quantization: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization

LLM Fundamentals

• What are foundation models, and how have they changed AI engineering?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• What is a Large Language Model (LLM), and how does it work?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization and How does an LLM work?
• Inside ChatGPT: What Happens After You Hit Enter?
  • Answer: Inside ChatGPT: What Happens After You Hit Enter
• What is the Transformer architecture and how does it work?
  • Answer: Decoding Transformer Architecture
• What are the key components of the Transformer architecture?
  • Answer: Decoding Transformer Architecture
• Walk me through what happens, step by step, in one forward pass of a decoder-only Transformer.
  • Answer: Decoding Transformer Architecture
• What is tokenization in LLMs?
  • Answer: Tokenization in LLMs and Tokenization in Large Language Models (LLMs)
• Explain BPE (Byte Pair Encoding).
  • Answer: Byte Pair Encoding
• Explain WordPiece and SentencePiece.
• What is positional encoding, and why is it needed in Transformers?
  • Answer: Positional Embeddings in LLMs
• What are embeddings?
  • Answer: Embeddings in Machine Learning
• Explain the Query(Q), Key(K), and Value(V) in attention.
  • Answer: Math behind Attention - Q, K, and V
• What is self-attention, and how does it work in Transformers?
  • Answer: Self Attention in Transformers
• What is Cross Attention in Transformers?
  • Answer: Cross Attention in Transformers
• Why do we scale the dot product attention by √dₖ in the Transformer architecture?
  • Answer: Math behind √dₖ Scaling Factor in Attention
• What is causal masking?
  • Answer: Causal Masking in Attention
• What are multi-head attention mechanisms? Why use multiple attention heads?
  • Answer: Multi-Head Attention in Transformers
• What are Feed-Forward Networks in LLMs?
  • Answer: Feed-Forward Networks in LLMs
• What is Generative AI?
  • Answer: What is Generative AI?
• What is the context window in LLMs, and why does it matter?
  • Answer: Context Window in LLMs and How does the Context Window work?
• Why is the context window limited in LLMs?
  • Answer: Why is the context window limited in LLMs?
• What is temperature in the context of LLMs, and how does it affect output?
  • Answer: How does Temperature control LLM output?
• Why is the first token slower than the rest in an LLM?
  • Answer: The First-Token Latency Problem in LLMs
• Explain Top-p (nucleus) sampling and Top-k sampling. How do they differ?
  • Answer: How do Top-k and Top-p Sampling work?
• Compare greedy decoding, beam search, top-k, top-p, and temperature sampling. When does each fail?
  • Answer: How do Top-k and Top-p Sampling work?
• What are logits, and how are they used in text generation?
  • Answer: Understanding Logits in Machine Learning and What are Logits in Machine Learning?
• What are stop tokens in LLMs, and how does an LLM know when to stop generating?
  • Answer: Stop Tokens in LLMs
• What is the difference between a stop token and a stop sequence?
  • Answer: Stop Tokens in LLMs
• What are skip connections (residual connections) in Transformers?
  • Answer: Skip connections (residual connections) in Transformers
• What is the difference between open-source and closed-source LLMs? When would you choose one over the other?
• What is the difference between encoder-only, decoder-only, and encoder-decoder Transformer architectures?
  • Answer: Encoder vs Decoder in Transformers
• What is KV cache, and how does it speed up inference?
  • Answer: What is KV Cache in LLMs?
• Estimate the KV cache memory needed to serve a large model. How does it constrain batch size and context length?
  • Answer: KV Cache Compression
• KV Cache Compression
  • Answer: KV Cache Compression
• What is model distillation, and how is it used with LLMs?
  • Answer: How does Knowledge Distillation work?
• What is Mixture of Experts (MoE), and how does it work in models like Mixtral?
  • Answer: Mixture of Experts Explained
• What is the difference between dense and sparse models?
  • Answer: Mixture of Experts Explained
• How does DeepSeek-V4 work?
  • Answer: DeepSeek-V4 Architecture Explained
• What is Flash Attention?
  • Answer: Decoding Flash Attention in LLMs
• What is Cross-Entropy Loss?
  • Answer: Math Behind Cross-Entropy Loss
• What is KL Divergence?
  • Answer: What is KL Divergence? The Math Explained Step by Step
• What is Grouped-Query Attention (GQA), and how does it differ from Multi-Head Attention (MHA)?
  • Answer: Grouped Query Attention
• How does Sliding Window Attention work?
  • Answer: How does Sliding Window Attention work?
• How do Attention Sinks work?
  • Answer: How do Attention Sinks work?
• How does Rotary Position Embedding (RoPE) work, and why is it preferred over learned positional embeddings?
  • Answer: Math Behind RoPE (Rotary Position Embedding)
• Explain Layer Normalization
  • Answer: Batch Normalization vs Layer Normalization
• Explain RMSNorm (Root Mean Square Layer Normalization)
  • Answer: RMSNorm (Root Mean Square Layer Normalization)
• Why do modern Transformers use Pre-LayerNorm (Pre-Norm) instead of Post-LayerNorm?
  • Answer: RMSNorm (Root Mean Square Layer Normalization)
• What are scaling laws (Chinchilla), and how do they guide model size vs training data decisions?
• Your LLM keeps ignoring your instructions. How do you make it follow structured output formats?
• Your LLM-powered tool hits the context window limit on long documents. How do you handle it?
• Your LLM does not admit when it does not know the answer. How do you make it say "I don't know"?
• Your LLM generates responses that are too verbose. How do you control response length?
• Your LLM memorized proprietary training data and leaks it in responses. How do you prevent this?
• Your LLM coding assistant generates outdated code using deprecated libraries. How do you fix it?
• Your tokenizer splits important domain terms into meaningless subword pieces. How do you fix it?
• Your Transformer's KV cache grows too large during long sequence generation. How do you manage memory?
  • Answer: Paged Attention in LLMs
• Your Transformer runs out of memory on long documents due to quadratic self-attention. How do you scale it?
  • Answer: How does Sliding Window Attention work? and Decoding Flash Attention in LLMs
• Your distilled student model fails on the complex reasoning that the teacher model handled. How do you close the gap?
• After RLHF alignment, your LLM became safer but lost capability on hard tasks. How do you manage the alignment tax?
• Your RLHF-trained LLM is gaming the reward model instead of being genuinely helpful. How do you fix reward hacking?
  • Answer: Reinforcement Learning from Human Feedback (RLHF)
• Your chatbot loses context after 10 turns in a conversation. How do you maintain a long conversation context?
  • Answer: AI Agent Memory
• Your chatbot fails when users switch topics mid-conversation. How do you handle topic switches?
• Your QA system always generates an answer even when no answer exists in the context. How do you detect unanswerable questions?
• Your summarization system hallucinated facts not in the original article. How do you fix it?
• Your text generation repeats phrases in long outputs. How do you fix repetition?
• Transformers work on text, so can they also understand images?
  • Answer: Decoding Vision Transformer (ViT)
• Small Language Models (SLMs)
  • Answer: Small Language Models (SLMs)
• Large Reasoning Models (LRMs)
  • Answer: Large Reasoning Models (LRMs)
• Jev and System One Models
  • Answer: Jev and System One Models Explained
• What are Autoregressive Models?
  • Answer: Autoregressive Models
• Explain the difference between autoregressive and masked language modeling.
• Proximal Policy Optimization (PPO)
  • Answer: Proximal Policy Optimization (PPO)
• Direct Preference Optimization (DPO)
  • Answer: Direct Preference Optimization (DPO)
• Group Relative Policy Optimization (GRPO)
  • Answer: Group Relative Policy Optimization (GRPO)
• Recursive Language Models (RLMs)
  • Answer: Recursive Language Models (RLMs)
• Continual Learning in LLMs
  • Answer: Continual Learning in LLMs
• What is Recursive Self-Improvement (RSI)?
  • Answer: What is Recursive Self-Improvement (RSI)?
• How do Diffusion Language Models (DLMs) work?
  • Answer: How do Diffusion Language Models (DLMs) work?
• How Does LLM Watermarking Work?
  • Answer: How Does LLM Watermarking Work?
• How do RNNs and Transformers differ?
  • Answer: How do RNNs and Transformers differ?

Prompt Engineering

• What is prompt engineering, and why is it critical for AI applications?
• Explain zero-shot, one-shot, and few-shot prompting with examples.
  • Answer: Explain zero-shot, one-shot, and few-shot prompting with examples
• What is chain-of-thought (CoT) prompting, and when should you use it?
  • Answer: How does Chain-of-Thought (CoT) Prompting work?
• Explain self-consistency prompting and how it improves reasoning.
• What is tree-of-thought prompting?
  • Answer: How does Tree of Thoughts work?
• What is ReAct (Reasoning + Acting) prompting, and how does it work?
  • Answer: ReAct Agent
• What is a system prompt, and how does it influence model behavior?
• How do you structure prompts for consistent structured output (JSON, XML)?
• What is prompt injection, and how do you defend against it?
  • Answer: Prompt Injection in LLMs
• What is jailbreaking in LLMs, and what are common jailbreak techniques?
  • Answer: Prompt Injection in LLMs and What is jailbreaking in LLMs?
• How do you optimize prompts for cost and latency?
• What is the difference between prompt engineering and prompt tuning?
• What is a prompt template, and how do you design one for production use?
• How do you handle multi-turn conversations with LLMs?
• What is role prompting, and when is it effective?
• What is prompt chaining, and how do you design a chain of prompts for complex tasks?
  • Answer: How does Prompt Chaining work?
• How do you evaluate and iterate on prompt quality?
• What are meta-prompts, and how can they be used to generate prompts?
• What are the common failure modes in prompting, and how do you debug them?
• How do you handle edge cases and adversarial inputs in prompt design?
• What is the "lost in the middle" problem in long-context prompting?
  • Answer: The Lost in the Middle Problem in LLMs
• What are output parsers, and why are they needed for production applications?
• How do you handle multi-language prompting effectively?
• Your few-shot prompting gives inconsistent results across similar inputs. How do you stabilize it?
• Your LLM classification system is too sensitive to prompt wording changes. How do you reduce prompt sensitivity?
• Your chatbot's system prompt containing proprietary business logic is being leaked by users. How do you prevent it?
• Your LLM agent is vulnerable to prompt injection that reveals the system prompt. How do you defend it?
  • Answer: Prompt Injection in LLMs
• Your chain-of-thought prompting is not improving LLM accuracy on reasoning tasks. What do you fix?
• Your AI system works in English but fails for other languages. How do you add multilingual support?
• Your zero-shot cross-lingual transfer from English fails on other languages. How do you fix it?

Retrieval-Augmented Generation (RAG)

• What is Retrieval-Augmented Generation (RAG), and why is it important?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• Explain the architecture of a basic RAG system.
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• What are the key components of a RAG pipeline?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• What are chunking strategies, and how do you choose the right chunk size?
  • Answer: Chunking Strategies for RAG
• Compare fixed-size chunking, semantic chunking, and recursive chunking.
  • Answer: Chunking Strategies for RAG
• What is Contextual Retrieval, and how does it improve RAG?
  • Answer: How does Contextual Retrieval work?
• What are embedding models, and how do they convert text to vectors?
  • Answer: What are Embeddings?
• How do you choose an embedding model for your RAG system?
• Explain Agentic RAG.
  • Answer: Agentic RAG
• What is hybrid search, and why is it better than pure vector search?
  • Answer: How does Hybrid Search work?
• What is re-ranking, and how does it improve RAG retrieval quality?
  • Answer: How does a Reranker work?
• What is ColBERT, and how does late interaction retrieval work?
  • Answer: ColBERT - Late Interaction Retrieval Explained
• Compare reranker architectures: cross-encoder, ColBERT, and LLM-based rerankers.
  • Answer: How does a Reranker work? and ColBERT - Late Interaction Retrieval Explained
• How do you handle multi-document and multi-hop questions in RAG?
  • Answer: Agentic RAG and GraphRAG
• What is the "lost in the middle" problem in RAG systems?
  • Answer: The Lost in the Middle Problem in LLMs
• How do you evaluate a RAG system? Explain faithfulness, relevance, and context precision/recall.
  • Answer: LLM Evaluation
• Explain Self-RAG. How does the model decide when to retrieve?
  • Answer: Agentic RAG
• What is GraphRAG, and when would you use it over traditional RAG?
  • Answer: GraphRAG
• Vectorless RAG
  • Answer: Vectorless RAG
• How do you handle structured data (tables, SQL databases) in a RAG pipeline?
• What are the common failure modes of RAG systems, and how do you debug them?
• How do you handle document updates and maintain freshness in a RAG system?
• How do you optimize RAG for latency in production?
• What is the role of metadata filtering in RAG systems?
  • Answer: How does a Vector Database work?
• Compare RAG vs fine-tuning. When would you use each?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• Long context windows keep getting cheaper. When should you use retrieval (RAG) vs putting everything in the context window?
  • Answer: The Lost in the Middle Problem in LLMs
• What is query transformation in RAG (HyDE, query decomposition, step-back prompting)?
  • Answer: How does HyDE work in RAG? and How do Query Rewriting and Multi-Query Retrieval work?
• How do you implement citation and source attribution in RAG?
• How do you scale a RAG system to millions of documents?
  • Answer: How does Approximate Nearest Neighbor (ANN) search work?
• What is parent-child chunking, and how does it improve retrieval?
  • Answer: Chunking Strategies for RAG
• Your RAG system is hallucinating despite having the right context. How do you fix it?
• Your RAG chunk overlap causes redundant results. How do you reduce redundancy?
• Your RAG retrieval is too slow with a large knowledge base. How do you speed it up?
  • Answer: How does Approximate Nearest Neighbor (ANN) search work?
• Your RAG system returns duplicate results. How do you deduplicate?
• Your RAG system needs per-user access control on internal documents. How do you implement it?
• Your RAG system fails on domain-specific jargon. How do you fix it?
  • Answer: How does Hybrid Search work?
• Your text-only RAG system now needs to handle images and tables. How do you extend it?
• Your RAG knowledge base gets updated frequently and needs versioning. How do you manage it?
• Your RAG system fails on multi-hop questions that require combining multiple facts. How do you fix it?
  • Answer: Agentic RAG and GraphRAG
• Your enterprise RAG system returns contradictory answers from different source documents. How do you resolve conflicts?
• Your RAG system returns outdated answers from an evolving knowledge base. How do you keep it current?
• Your RAG system struggles with PDF documents containing tables and layouts. How do you fix PDF parsing?

AI Agents and Agentic Systems

• What is an AI agent, and how does it differ from a simple LLM call?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization and AI Agent Explained
• AI Agent Memory
  • Answer: AI Agent Memory
• Harness Engineering in AI
  • Answer: Harness Engineering in AI
• What matters more for an agentic coding tool like Claude Code: the model or the harness?
  • Answer: Harness Engineering in AI and How does Claude Code work?
• Explain the ReAct (Reasoning + Acting) agent architecture.
  • Answer: ReAct Agent
• What is the Plan-and-Execute agent pattern?
  • Answer: Plan-and-Execute Agent
• What is tool use (function calling) in LLMs, and how does it enable agents?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization and How does Function Calling work in LLMs?
• What is the difference between structured output and function calling?
  • Answer: How does Function Calling work in LLMs?
• How do you design and define tools for an AI agent?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• How does an agent decide when to call a tool versus answering from its own knowledge?
• What is the difference between single-agent and multi-agent systems?
  • Answer: Multi-Agent Systems
• When do multi-agent systems break down, and when is a single agent the better choice?
  • Answer: Multi-Agent Systems
• What is Model Context Protocol (MCP), and how does it standardize tool integration?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization and What is MCP (Model Context Protocol)?
• How does MCP differ from traditional function calling?
  • Answer: What is MCP (Model Context Protocol)? and How does Function Calling work in LLMs?
• What are AI SubAgents?
  • Answer: AI SubAgents
• What are the different types of agent memory (short-term, long-term, episodic)?
  • Answer: AI Agent Memory
• How do you handle agent failures and implement error recovery?
• What is an agent loop, and how does it decide when to stop?
  • Answer: AI Agent Loop
• Context Engineering
  • Answer: Context Engineering
• How does context compaction work?
  • Answer: How does context compaction work?
• Loop Engineering
  • Answer: Loop Engineering
• Graph Engineering
  • Answer: Graph Engineering
• How AI Agents Communicate?
  • Answer: How AI Agents Communicate
• What are Agent Skills?
  • Answer: What are Agent Skills?
• How do you evaluate and test AI agents?
  • Answer: AI Agent Evaluation
• What are the security risks of agentic systems, and how do you mitigate them?
  • Answer: Prompt Injection in LLMs
• Your agent reads untrusted content (emails, web pages, documents) and can call tools. How do you prevent indirect prompt injection and data exfiltration?
  • Answer: Prompt Injection in LLMs
• What is the difference between reactive and proactive agents?
• How do you manage token consumption and cost in long-running agent workflows?
  • Answer: How does context compaction work? and How would you reduce the token consumption?
• What is the human-in-the-loop pattern for agents, and when is it needed?
• How do you implement guardrails for AI agents to prevent harmful actions?
  • Answer: How do LLM guardrails work?
• What is agent reflection, and how does it improve agent performance?
  • Answer: Reflection Agent
• What is the difference between code-generating agents and tool-calling agents?
• How do you handle multi-modal inputs and outputs in agentic systems?
• How do you implement state management in complex agent workflows?
  • Answer: How does LangGraph work?
• How do you build a customer support agent with escalation logic?
• What is agent orchestration, and how do you implement it?
  • Answer: AI Orchestration
• What is Sakana Fugu, and how does it orchestrate a team of AI models?
  • Answer: Sakana Fugu - The Technical Report Explained
• How do you build a code execution agent safely using sandboxed environments?
• Your AI agent is stuck in an infinite loop. How do you detect and break the cycle?
  • Answer: Fix an infinite loop in an AI agent
• Your AI agent gets conflicting answers from different tools. How does it reconcile them?
• Your AI agent burns too many tokens per task. How do you reduce token consumption?
  • Answer: How would you reduce the token consumption?
• Your AI agent keeps exceeding its budget per task. How do you enforce budget limits?
  • Answer: AI Agent Loop
• Your AI agent hallucinates tool capabilities and passes wrong inputs. How do you fix it?
• Your AI agent deleted a production database. How do you prevent irreversible actions?
• Your AI agent has many tools, but keeps picking the wrong one. How do you improve tool selection?
• Your AI agent takes too long to complete a task. How do you speed it up?
• Your long-running agent drifts after hours and confidently works on the wrong thing. How do you diagnose and fix it?
• Your LLM selects the right tool but extracts the wrong parameters. How do you fix parameter extraction?
• How do Computer-Use Agents work?
  • Answer: How do Computer-Use Agents work?
• How do Browser Agents work?
  • Answer: How do Browser Agents work?
• How does LangChain work?
  • Answer: How does LangChain work?
• How does LangGraph work?
  • Answer: How does LangGraph work?
• What is OKF (Open Knowledge Format)?
  • Answer: What is OKF (Open Knowledge Format)?

Fine-Tuning and Model Adaptation

• What is fine-tuning, and when should you fine-tune an LLM?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• Explain the difference between full fine-tuning and parameter-efficient fine-tuning (PEFT).
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• What is LoRA (Low-Rank Adaptation), and how does it work?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• What is QLoRA, and how does it enable fine-tuning on consumer hardware?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• How does fine-tuning work?
  • Answer: How does fine-tuning work?
• Explain Prefix Tuning and Prompt Tuning. How are they different from LoRA?
  • Answer: How does Prefix Tuning work?
• What is adapter-based fine-tuning?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• Explain the difference between pre-training, supervised fine-tuning (SFT), and preference optimization (RLHF/DPO).
  • Answer: Decoding InstructGPT and How does Pretraining work?
• What is RLHF (Reinforcement Learning from Human Feedback), and how is it used to align LLMs?
  • Answer: Reinforcement Learning from Human Feedback (RLHF)
• What is Deep RL from Human Preferences, the paper that started RLHF?
  • Answer: Deep RL from Human Preferences
• Why did DPO displace PPO-based RLHF at many labs? When is online RL still better?
  • Answer: Direct Preference Optimization (DPO) and Proximal Policy Optimization (PPO)
• What is instruction tuning, and why is it important for chat models?
  • Answer: Decoding InstructGPT
• How do you prepare a dataset for fine-tuning an LLM?
• What is catastrophic forgetting, and how do you prevent it during fine-tuning?
  • Answer: Continual Learning in LLMs
• When should you choose fine-tuning over RAG over prompt engineering?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization
• How do you evaluate a fine-tuned model's performance?
  • Answer: LLM Evaluation
• What is synthetic data generation, and how do you use it for fine-tuning?
• What are the key hyperparameters for fine-tuning (learning rate, epochs, batch size, LoRA rank)?
  • Answer: LoRA - Low-Rank Adaptation of LLMs
• Do the GPU memory math for full fine-tuning a 7B model in bf16 with Adam. Now with LoRA and QLoRA.
• How do you fine-tune a model for a specific domain (legal, medical, finance)?
• What is continual pre-training, and when would you use it?
• How do you merge multiple LoRA adapters?
  • Answer: LoRA - Low-Rank Adaptation of LLMs
• What is the difference between SFT (Supervised Fine-Tuning) and alignment training?
  • Answer: Decoding InstructGPT
• What is RLAIF (RL from AI Feedback), and how does it differ from RLHF?
• What is Constitutional AI, and how does it differ from RLHF?
• What is RLVR (Reinforcement Learning with Verifiable Rewards), and when does it beat a learned reward model?
  • Answer: Group Relative Policy Optimization (GRPO)
• What is knowledge distillation for fine-tuning, and what are the legal considerations?
  • Answer: How does Knowledge Distillation work?
• Your fine-tuned LLM produces factually wrong outputs due to training data quality issues. How do you fix it?
• You must choose between LoRA and full fine-tuning for a domain-specific assistant. How do you decide?
  • Answer: LoRA - Low-Rank Adaptation of LLMs
• Your fine-tuned model memorized training data verbatim instead of learning patterns. How do you fix overfitting?
• Your fine-tuned LLM forgot its general capabilities after domain-specific fine-tuning. How do you fix catastrophic forgetting?
  • Answer: Continual Learning in LLMs
• Your RLHF preference data has low annotator agreement. How do you ensure data quality?

Vector Databases and Embeddings

• What are embeddings in the context of AI engineering?
  • Answer: Embeddings in Machine Learning
• How do embedding models convert text to vectors?
  • Answer: What are Embeddings?
• What is Contrastive Learning, and how is it used to train embedding models?
  • Answer: What is Contrastive Learning?
• What is the difference between sparse and dense embeddings?
• Explain cosine similarity, dot product, and Euclidean distance for vector search.
  • Answer: How does a Vector Database work? and How does Cosine Similarity work?
• What is a vector database, and how does it differ from a traditional database?
  • Answer: How does a Vector Database work?
• How does Approximate Nearest Neighbor (ANN) search work?
  • Answer: How does Approximate Nearest Neighbor (ANN) search work?
• Compare HNSW, IVF, and flat indexes. How do you pick one, and what does recall@k cost in latency?
  • Answer: How does Approximate Nearest Neighbor (ANN) search work?
• How does an Embedding Cache work?
  • Answer: How does an Embedding Cache work?
• How do you choose the right embedding model for your use case?
• What is embedding dimensionality, and how does it affect performance and cost?
• How do you handle embedding drift when the embedding model is updated?
• What are multi-modal embeddings, and how are they generated?
  • Answer: Multimodal AI
• How do you index and query multi-tenant data in a vector database?
• What is quantization of embeddings, and how does it reduce storage costs?
• How do you benchmark and evaluate embedding model quality?
• What is the role of metadata in vector databases?
  • Answer: How does a Vector Database work?
• How do you handle large-scale vector search with billions of vectors?
  • Answer: How does Approximate Nearest Neighbor (ANN) search work?
• What is hybrid search (combining keyword search with vector search)?
  • Answer: How does Hybrid Search work?
• How do you fine-tune an embedding model for a specific domain?
  • Answer: What is Contrastive Learning?
• Your vector database for RAG is consuming too much memory. How do you reduce it?
• Your vector database cannot scale to millions of embeddings. How do you fix the bottleneck?
  • Answer: How does Approximate Nearest Neighbor (ANN) search work?
• Your new embedding model has different dimensions from the existing vectors in production. How do you handle the mismatch?
• Your vector search returns irrelevant results despite high similarity scores. How do you fix it?
• You deployed a new embedding model, and search quality crashed overnight. How do you handle embedding drift?
• Your semantic search fails for short queries. How do you improve it?

AI System Design

• Design a Real-Time Voice AI Agent
  • Answer: Design a Real-Time Voice AI Agent
• Design ChatGPT: Training to Serving (End to End)
• Design a RAG System (Chat with Your Documents)
• Design an enterprise RAG assistant over 10M documents with per-user permissions.
• Design Memory for a Personal AI Assistant
  • Answer: AI Agent Memory and How does ChatGPT remember you?
• Design a Deep Research Agent
• Design a Multi-Agent Customer Support System
  • Answer: Multi-Agent Systems
• Design an On-Device AI Assistant
  • Answer: Cloud vs On-Device Model Deployment and How does llama.cpp run LLMs on everyday hardware?
• Design a Multimodal Search System (Text, Image, Video)
• Design an LLM Inference Platform (vLLM-as-a-Service)
  • Answer: How does vLLM work? and LLM Inference Optimization
• Design an LLM Evaluation Platform
  • Answer: LLM Evaluation
• Design a Text-to-Image Generation Service (Midjourney-like)
• Design a Music Generation Service (Suno-like)
• Design a Video Generation Service (Sora-like)
• Design an AI Coding Agent.
  • Answer: How does Claude Code work? and How does Cursor work?
• Design a code generation and review system.
• Design a content moderation system using AI.
• Design a real-time AI recommendation system.
• Design an AI-powered email assistant.
• Design a medical diagnosis assistant using AI.
• Design a fraud detection system powered by LLMs.
• Design an AI-powered data extraction pipeline from unstructured documents.
• Design a Text-to-SQL system over a data warehouse with thousands of tables.
• Design a personalized learning assistant.
• Design an AI system for automated code migration.
• Design an AI-powered legal document review system.
• Design a conversational AI system with memory across sessions.
  • Answer: AI Agent Memory and How does ChatGPT remember you?
• How do you design for latency vs quality trade-offs in AI systems?
• How do you implement caching strategies for LLM applications?
  • Answer: How does Prompt Caching work? and How does Semantic Caching work?
• How do you design rate limiting and cost management for AI APIs?
• How do you handle failover and fallback strategies for AI systems?
• How do you design an AI system for high availability and fault tolerance?
• How do you design an AI system that gracefully degrades when the model is unavailable?
• What are the key considerations for multi-region deployment of AI systems?
• Design an AI-powered search engine for an e-commerce platform.
• Design an AI gateway/proxy for managing LLM access across an organization.
• How do you design a RAG system that handles conflicting information across sources?
• How do you approach capacity planning for an AI system?
• Design a multi-tenant AI chatbot platform where each business gets a custom chatbot.
• Design an AI meeting summarizer system for thousands of meetings daily.
• Design an AI notification system that prioritizes instead of broadcasting.
• Design an AI-powered anomaly detection system for cloud infrastructure.
• Design an AI-powered document processing pipeline for financial institutions.
• Design an AI dynamic pricing engine.
• Design an AI resume screening system that handles 100K applications per week.
• Design an AI voice assistant architecture.
  • Answer: Design a Real-Time Voice AI Agent
• Design a multi-agent workflow system where agents collaborate on complex tasks.
  • Answer: Multi-Agent Systems
• Design a real-time AI transcription system for concurrent audio streams.
• Design an AI-powered live streaming content moderation system.

LLMOps and Production AI

• How does Prompt Caching work?
  • Answer: How does Prompt Caching work?
• Prefill vs Decode
  • Answer: Prefill vs Decode: LLM Inference Optimization
• Why is prefill compute-bound and decode memory-bandwidth-bound?
  • Answer: Prefill vs Decode: LLM Inference Optimization
• What is chunked prefill, and why does it improve tail latency under mixed traffic?
  • Answer: Prefill vs Decode: LLM Inference Optimization
• What is Prefill-Decode Disaggregation, and when does it pay off?
  • Answer: Prefill-Decode Disaggregation in LLM Inference
• Explain the AI product lifecycle from ideation to production.
• What is LLMOps, and how does it differ from traditional MLOps?
• How do you serve LLMs in production?
  • Answer: How does vLLM work?, LLM Inference Optimization and How does Model Serving work?
• What is model quantization?
  • Answer: Explained in this video: AI Engineering Explained: LLM, RAG, MCP, Agent, Fine-Tuning, Quantization and How does Model Quantization work?
• Explain post-training quantization (PTQ) vs quantization-aware training (QAT). What breaks when you push weights to 2-4 bits?
  • Answer: How does Model Quantization work?
• How do you monitor LLM applications in production?
  • Answer: AI Agent Observability
• What is LLM observability?
  • Answer: AI Agent Observability
• What are guardrails for LLMs, and how do you implement them?
  • Answer: How do LLM guardrails work?
• How do you implement content filtering for AI outputs?
  • Answer: How do LLM guardrails work?
• How do you estimate the cost of running an AI-powered feature in production?
• How do you optimize LLM inference costs in production?
  • Answer: LLM Inference Optimization
• How do you implement A/B testing for LLM systems?
• What is CI/CD for AI applications, and how does it differ from traditional CI/CD?
• How do you version and manage prompts in production?
• What is model versioning, and how do you handle model rollbacks?
• How do you implement rate limiting and throttling for LLM APIs?
• How do you handle model updates and migrations without downtime?
• What is the role of feature flags in AI deployments?
• How do you implement logging and tracing for LLM applications?
  • Answer: AI Agent Observability
• How do you handle PII and sensitive data in LLM inputs and outputs?
• What is a gateway pattern for LLM API management?
• How does Token Streaming work?
  • Answer: How does Token Streaming work?
• How do you implement streaming responses for real-time AI applications?
  • Answer: How does Token Streaming work?
• How does vLLM work?
  • Answer: How does vLLM work?
• How does SGLang work?
  • Answer: How does SGLang work?
• How does TensorRT-LLM work?
  • Answer: How does TensorRT-LLM work?
• How does llama.cpp run LLMs on everyday hardware?
  • Answer: How does llama.cpp run LLMs on everyday hardware?
• How does Ollama work?
  • Answer: How does Ollama work?
• When would you choose vLLM vs SGLang vs TensorRT-LLM?
  • Answer: How does vLLM work?, How does SGLang work? and How does TensorRT-LLM work?
• What are the key SLAs and metrics for production AI systems (latency, throughput, availability)?
• Cloud vs on-device Model Deployment for AI applications.
  • Answer: Cloud vs On-Device Model Deployment
• How do you implement fallback strategies when the primary model is unavailable or rate-limited?
• How do you implement structured output from LLMs reliably in production?
  • Answer: How does Function Calling work in LLMs?
• How do you handle long contexts efficiently in production (context compression, prefix caching)?
  • Answer: How does Prompt Caching work? and How does context compaction work?
• What is semantic routing, and how do you implement it in a multi-model system?
  • Answer: LLM Routing
• How do you manage secrets and API keys securely in LLM applications?
• Your LLM API has latency spikes during peak hours. How do you stabilize it?
• Your LLM endpoint's p99 latency doubled after a deploy with no model change. How do you diagnose it?
• Your LLM costs are too high in production. How do you reduce costs without degrading quality?
  • Answer: LLM Routing and How does Semantic Caching work?
• Your application is hitting LLM provider rate limits during peak hours. How do you handle it?
• Your application depends on one LLM provider. How do you switch providers without downtime?
• Your AI system handles 100 requests/sec but crashes at 5000. How do you scale for concurrent requests?
• A traffic spike brings down your AI system. How do you handle peak traffic?
• One LLM provider outage took down your entire system. How do you eliminate single points of failure?
• Your multi-LLM pipeline fails when one model in the chain breaks. How do you handle orchestration failure?
  • Answer: AI Orchestration
• Your AI pipeline has zero visibility into which step is failing. How do you add observability?
  • Answer: AI Agent Observability
• You quantized your LLM, but accuracy dropped significantly. How do you minimize quantization loss?
  • Answer: How does Model Quantization work?
• One failing AI component can take down your entire platform. How do you design graceful degradation?

Evaluation and Testing

• AI Agent Evaluation
  • Answer: AI Agent Evaluation
• LLM Evaluation
  • Answer: LLM Evaluation
• AI Agent Observability
  • Answer: AI Agent Observability
• What is evaluation-driven development for AI applications?
• Why is AI only as good as our definition of done?
  • Answer: AI Is Only as Good as Our Definition of Done
• How do you evaluate LLM outputs? What metrics do you use?
  • Answer: LLM Evaluation
• Explain BLEU, ROUGE, and BERTScore. When would you use each?
  • Answer: LLM Evaluation
• What is G-Eval, and how does it use LLMs for evaluation?
  • Answer: LLM as a Judge
• What is LLM-as-a-judge evaluation, and what are its limitations?
  • Answer: LLM as a Judge
• How do you conduct human evaluation for AI systems?
  • Answer: LLM Evaluation
• What is red teaming, and how do you red team an LLM application?
• How do you detect and measure hallucinations in LLM outputs?
• What is adversarial testing for AI systems?
• How do you build a regression test suite for AI applications?
• What are benchmark suites (MMLU, HumanEval, GSM8K), and how do you interpret them?
  • Answer: LLM Evaluation
• What is benchmark contamination, and how do you guard against it?
  • Answer: LLM Evaluation
• Your new model version scores higher on every benchmark, but users say it got worse. Why does this happen, and how do you find the problem?
• How do you evaluate a RAG system end-to-end?
  • Answer: LLM Evaluation
• How do you evaluate the quality of AI agents?
  • Answer: AI Agent Evaluation
• How would you evaluate an autonomous coding agent? Why can SWE-bench pass rates be misleading?
  • Answer: AI Agent Evaluation
• What is the difference between offline and online evaluation for AI systems?
• How do you measure factual consistency in LLM outputs?
• How do you evaluate multi-turn conversation quality?
• What is the role of golden datasets in AI evaluation?
• How do you build an eval set when there is no labelled ground truth and domain experts are expensive?
• How do you implement continuous evaluation for production AI systems?
• How do you evaluate bias in AI model outputs?
• How do you compare two models or prompts in a statistically rigorous way?
• How do you evaluate the robustness of an LLM application across input variations?
• What are the key differences between evaluating traditional ML vs LLM applications?
• How do you set up an evaluation framework from scratch for a new LLM application?
  • Answer: LLM Evaluation
• Your model passes one fairness metric but fails another. How do you handle conflicting audit results?
• Your model was fair at deployment, but became biased 6 months later. How do you monitor continuously?
• An external auditor cannot reproduce your model's results. How do you ensure audit reproducibility?
• How do you structure red teaming for an LLM chatbot before launch?
• How do you red team a multimodal model where text-only safety tests miss cross-modal attacks?

AI Safety, Ethics, and Responsible AI

• What are hallucinations in LLMs, and how do you mitigate them?
  • Answer: Why Do LLMs Hallucinate? The Token-Level Explanation
• What is prompt injection, and what are the different types (direct, indirect)?
  • Answer: Prompt Injection in LLMs
• How do you implement input and output guardrails for AI systems?
  • Answer: How do LLM guardrails work?
• What is AI alignment, and why is it important?
  • Answer: Decoding InstructGPT and Reinforcement Learning from Human Feedback (RLHF)
• How do you detect and mitigate bias in AI systems?
• What are the key data privacy considerations (GDPR, CCPA) when building AI applications?
• How do you handle PII in LLM inputs and outputs?
• What is explainability in AI, and why does it matter?
• What is the difference between interpretability and explainability?
• What is mechanistic interpretability, and why do AI labs invest in it?
• How