Interview Q&A · 50 questions · 12 answered publicly
Native AI Engineering Interview Questions
Ordered by how often they're asked. Sign in free to see every answer, rate your confidence and run timed mock interviews.
-
Most asked What is RAG?
Summary
Retrieval-Augmented Generation grounds LLM answers in fetched documents to reduce hallucination and use private data.
Longer Answer
Retrieval-Augmented Generation grounds LLM answers in fetched documents to reduce hallucination and use private data. RAG / Native AI engineering.
Key Points
- Retrieval-Augmented Generation grounds LLM answers in fetched documents to reduce hallucination and use private data
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'What is RAG?'
Comparison / Tradeoffs
Aspect Notes Core idea Retrieval-Augmented Generation grounds LLM answers in fetched documents to reduce hallucination and use private data. Practical tradeoff RAG / Native AI engineering. -
Most asked Explain: RAG pipeline stages.
Summary
Ingest → chunk → embed → index → retrieve → (re)rank → prompt → generate → cite.
Longer Answer
Ingest → chunk → embed → index → retrieve → (re)rank → prompt → generate → cite. RAG / Native AI engineering.
Key Points
- Ingest → chunk → embed → index → retrieve → (re)rank → prompt → generate → cite
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: RAG pipeline stages.'
Comparison / Tradeoffs
Aspect Notes Core idea Ingest → chunk → embed → index → retrieve → (re)rank → prompt → generate → cite. Practical tradeoff RAG / Native AI engineering. -
Most asked Explain: chunking strategies.
Summary
Fixed size, recursive, semantic, or document-structure aware; overlap to preserve context.
Longer Answer
Fixed size, recursive, semantic, or document-structure aware; overlap to preserve context. RAG / Native AI engineering.
Key Points
- Fixed size, recursive, semantic, or document-structure aware; overlap to preserve context
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: chunking strategies.'
Comparison / Tradeoffs
Aspect Notes Core idea Fixed size, recursive, semantic, or document-structure aware; overlap to preserve context. Practical tradeoff RAG / Native AI engineering. -
Most asked Explain: hybrid search.
Summary
Combine BM25 keyword with vector search for better recall/precision.
Longer Answer
Combine BM25 keyword with vector search for better recall/precision. RAG / Native AI engineering.
Key Points
- Combine BM25 keyword with vector search for better recall/precision
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: hybrid search.'
Comparison / Tradeoffs
Aspect Notes Core idea Combine BM25 keyword with vector search for better recall/precision. Practical tradeoff RAG / Native AI engineering. -
Most asked Explain: RAG failure modes.
Summary
Bad chunks, weak retrieval, prompt stuffing noise, stale indexes, ACL leaks.
Longer Answer
Bad chunks, weak retrieval, prompt stuffing noise, stale indexes, ACL leaks. RAG / Native AI engineering.
Key Points
- Bad chunks, weak retrieval, prompt stuffing noise, stale indexes, ACL leaks
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: RAG failure modes.'
Comparison / Tradeoffs
Aspect Notes Core idea Bad chunks, weak retrieval, prompt stuffing noise, stale indexes, ACL leaks. Practical tradeoff RAG / Native AI engineering. -
Most asked Explain: hallucination mitigation.
Summary
Grounding, refusals when evidence missing, lower temperature, tool checks.
Longer Answer
Grounding, refusals when evidence missing, lower temperature, tool checks. RAG / Native AI engineering.
Key Points
- Grounding, refusals when evidence missing, lower temperature, tool checks
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: hallucination mitigation.'
Comparison / Tradeoffs
Aspect Notes Core idea Grounding, refusals when evidence missing, lower temperature, tool checks. Practical tradeoff RAG / Native AI engineering. -
Most asked Agents: tool calling.
Summary
Model emits structured function calls; host executes and returns results.
Longer Answer
Model emits structured function calls; host executes and returns results. Agentic systems.
Key Points
- Model emits structured function calls; host executes and returns results
- Agentic systems
- Relevant areas: Ai, Agents
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Agents: tool calling.'
Comparison / Tradeoffs
Aspect Notes Core idea Model emits structured function calls; host executes and returns results. Practical tradeoff Agentic systems. -
Most asked LLMOps: fine-tuning vs RAG.
Summary
RAG for knowledge; fine-tune for style/format/tool behavior.
Longer Answer
RAG for knowledge; fine-tune for style/format/tool behavior. LLMOps practice.
Key Points
- RAG for knowledge; fine-tune for style/format/tool behavior
- LLMOps practice
- Relevant areas: Ai, Llmops
- Use the answer to connect the concept back to a realistic production or interview scenario for 'LLMOps: fine-tuning vs RAG.'
Comparison / Tradeoffs
Aspect Notes Core idea RAG for knowledge; fine-tune for style/format/tool behavior. Practical tradeoff LLMOps practice. -
Most asked LLMOps: prompt injection.
Summary
Untrusted content instructing the model; isolate and sanitize.
Longer Answer
Untrusted content instructing the model; isolate and sanitize. LLMOps practice.
Key Points
- Untrusted content instructing the model; isolate and sanitize
- LLMOps practice
- Relevant areas: Ai, Llmops
- Use the answer to connect the concept back to a realistic production or interview scenario for 'LLMOps: prompt injection.'
Comparison / Tradeoffs
Aspect Notes Core idea Untrusted content instructing the model; isolate and sanitize. Practical tradeoff LLMOps practice. -
Most asked Explain: vector similarity.
Summary
Cosine/dot/L2 metrics; normalize embeddings when using cosine/dot.
Longer Answer
Cosine/dot/L2 metrics; normalize embeddings when using cosine/dot. RAG / Native AI engineering.
Key Points
- Cosine/dot/L2 metrics; normalize embeddings when using cosine/dot
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: vector similarity.'
Comparison / Tradeoffs
Aspect Notes Core idea Cosine/dot/L2 metrics; normalize embeddings when using cosine/dot. Practical tradeoff RAG / Native AI engineering. -
Most asked Explain: reranking.
Summary
Cross-encoder or LLM rerank top-k to improve precision before generation.
Longer Answer
Cross-encoder or LLM rerank top-k to improve precision before generation. RAG / Native AI engineering.
Key Points
- Cross-encoder or LLM rerank top-k to improve precision before generation
- RAG / Native AI engineering
- Relevant areas: Ai, Rag
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Explain: reranking.'
Comparison / Tradeoffs
Aspect Notes Core idea Cross-encoder or LLM rerank top-k to improve precision before generation. Practical tradeoff RAG / Native AI engineering. -
Most asked Agents: ReAct pattern.
Summary
Interleave reasoning traces with actions/observations.
Longer Answer
Interleave reasoning traces with actions/observations. Agentic systems.
Key Points
- Interleave reasoning traces with actions/observations
- Agentic systems
- Relevant areas: Ai, Agents
- Use the answer to connect the concept back to a realistic production or interview scenario for 'Agents: ReAct pattern.'
Comparison / Tradeoffs
Aspect Notes Core idea Interleave reasoning traces with actions/observations. Practical tradeoff Agentic systems. - Explain: embedding models.
- Explain: metadata filtering.
- Explain: citation / grounding.
- Explain: context window limits.
- Explain: evaluation of RAG.
- Explain: security in RAG.
- What is an LLM agent?
- Agents: when not to use agents.
- Agents: guardrails.
- Agents: infinite loops.
- LLMOps: prompt versioning.
- LLMOps: eval harness.
- LLMOps: observability for LLMs.
- LLMOps: cost control.
- LLMOps: safety filters.
- LLMOps: PII handling.
- LLMOps: structured output.
- Explain: index freshness.
- Explain: query rewriting.
- Explain: parent-document retrieval.
- Agents: planner–executor.
- Agents: multi-agent systems.
- Agents: agent memory.
- Agents: deterministic vs stochastic tools.
- LLMOps: online evaluation.
- LLMOps: A/B testing prompts.
- LLMOps: latency budgets.
- LLMOps: model routing.
- LLMOps: fallback strategies.
- LLMOps: data retention.
- LLMOps: rate limits / retries.
- LLMOps: jailbreak robustness.
- Explain: HyDE.
- Explain: graph RAG.
- Explain: multimodal RAG.
- LLMOps: LoRA / PEFT.
- LLMOps: distillation.
- LLMOps: batch inference.