Articles

What GenAI and LLM engineer interviews actually ask, grouped by topic, with what a strong answer sounds like and what a weak one gives away.

October 13, 2026

Imagine an interviewer asks how to adapt Amazon Nova 2 Lite for a specialist domain: should you feed it a vast library of raw documents, or show it carefully…

October 10, 2026

Each lifecycle stage by what it produces, the two feedback loops around deployment, which AWS service fits each stage, and the seven things MLOps adds.

October 8, 2026

A manager asks for "an AI solution" to calculate shipping fees.

October 8, 2026

Two layers interviewers want to hear: IAM, TLS, CloudTrail and encryption to protect the system, and Bedrock Guardrails grounding checks to block hallucinations.

October 7, 2026

Bias means two things on AIF-C01: a model too simple to fit, or unfair results for one group. How to tell which, and when to use Clarify, Model Monitor or A2I.

October 2, 2026

Bedrock gives you managed foundation models behind an API; SageMaker AI gives you the full ML platform. Four questions that tell you which one your project needs.

September 30, 2026

Tokens are what a foundation model reads, embeddings make meaning comparable, and the lifecycle turns a base model into a solution. How AIF-C01 tests all three.

September 29, 2026

Almost every team iterates on prompts. Almost none of them evaluate the result. Here is why eyeballing a few examples is not evidence, and what the smallest honest alternative looks like.

August 27, 2026

A grounding score turns 'the chatbot made something up' from a complaint into a number you can watch. Here is what it measures, what it cannot see, and where it belongs in a production retrieval system.

August 27, 2026

Most people preparing for AI engineering roles do not have a material shortage — they have a sequencing problem. Here is what unordered learning actually costs, and what a sequenced path replaces it with.

August 25, 2026

Definition questions have quietly disappeared from AI engineering interviews. What replaced them measures something a textbook cannot give you — and here is what interviewers are listening for instead.

August 25, 2026

"How does an LLM generate a response?" sounds like a warm-up question. It isn't. Here's why most answers stall after two beats, and what the full answer actually requires.

July 19, 2026

LLM APIs bill per token, and technical text can take 50% more tokens per word than plain English. The simple formula to price a prompt before it ships.

July 19, 2026

Interviewers ask one open-ended question about production prompt design and listen for five specific things. Here's what those five things actually are, and which one most candidates skip.

July 19, 2026

RCTF stands for Role, Context, Task, Format: the four parts of a system prompt that holds up in production. Skip Format and output drifted in 8 of 10 runs.

July 19, 2026

Interview questions about tool use and agent loops aren't really testing whether you can name the steps — they're testing whether you've watched one fail. Here's what the questions are really probing for.

July 19, 2026

Every AI agent that takes real actions relies on a boundary most people never name explicitly: the model requests, your code decides. Here's why that boundary is the whole design, and what breaks when it's missing.

July 19, 2026

A single AI agent fails loudly. A team of them can fail silently — passing garbage from one agent to the next while reporting success. Here's the design pattern that catches both failure modes before production does.

July 19, 2026

Multi-agent system design has become one of the default AI engineering interview topics. The question isn't whether you can name the pattern — it's whether you can explain what breaks it.

July 19, 2026

Production AI safety interview questions have a shape: six specific beats, and most candidates only cover four of them. Here's what the missing two reveal about your answer.

July 19, 2026

Guardrails, evals, and observability aren't three separate projects — they're three layers of the same safety stack, and most teams only build the first one. Here's why that's not enough.

July 19, 2026

Name the one constraint that can't flex, then pick the tier. On one workload a frontier model cost about 20x a lightweight one for the same requests.

July 19, 2026

Interviewers don't want a model name. They want to hear you name a constraint first. Here's why that one habit is the whole signal — and what it reveals about fine-tuning, benchmarks, and production trade-offs.

July 19, 2026

A career-switcher guide to AI portfolio projects: how engineers, data analysts, PMs, and ML-adjacent professionals can choose projects that make past experience relevant.

July 19, 2026

A practical AI/ML resume project list: RAG, agents, evals, streaming APIs, SageMaker, monitoring, and other portfolio builds hiring teams can inspect.

July 19, 2026

A practical roadmap for building an AI portfolio from zero experience: start with small LLM tools, add RAG, evals, deployment, and a public project narrative.

July 19, 2026

An LLM remembers nothing between calls. Keep a rolling summary, pin key facts, and retrieve old turns on demand. Costs, a worked example and a fix list.

July 15, 2026

SageMaker serves models four ways: real-time, serverless, asynchronous and batch. Their limits, what each costs while idle, and a deploy-and-clean-up script.

June 14, 2026

Most teams fine-tune too early. The 3 questions to ask first, what each option really costs, a decision checklist, worked examples and a Python benchmark.

June 10, 2026

Retrieval-Augmented Generation is everywhere. But most tutorials skip the tradeoffs. Here's what actually matters when you're building production RAG.

June 8, 2026

ML teams consistently over-permission their AWS roles. Here's how execution roles actually work, what permissions SageMaker jobs actually need, and a Terraform module to create it right.

June 5, 2026

Certificates don't get you hired. Deployed projects do. Here's the difference between passive and active learning in AI.

June 1, 2026