Introduction to AI Engineering

If you're interviewing or transitioning into an AI engineering role, this course is for you.

Generative AI models have spawned a new type of engineering: the AI Engineer. Sometimes called applied AI engineering or agentic engineering, this role spans machine learning, infrastructure, and full-stack product engineering. As an AI engineer, you're primarily responsible for turning AI models into products: agents, retrieval systems, and evaluation pipelines, powering everything from coding assistants to AI-powered customer support.

The interview process for this role is evolving quickly, too. This course covers the latest interview loops at frontier AI labs and startups—including Anthropic, OpenAI, Google, and Meta—and dives into real questions that interviewers are asking, including newer interviews on agentic system design, LLM fundamentals, and practical, AI-enabled coding problems.

Goals of this course

By the end of this course, you should be able to:

• Understand the current AI engineer interview loops at top companies and AI labs
• Answer system design questions, including traditional and newer agentic systems
• Speak fluently on GenAI concepts like prompt engineering, RAG, MCP, and evals
• Complete practical and algorithmic coding problems, with or without AI assistance
• Tell strong behavioral stories that draw on your experience leading AI projects

What is an applied AI or agentic engineer?

Generally speaking, an AI engineer is a software engineer that builds systems using large language models. While AI/ML researchers develop new foundation models, an AI engineer builds systems around them that transform raw model outputs into reliable products and results.

"Agentic engineer" is the latest variant of AI engineering, and it reflects how the role is continuing to evolve. The first wave of LLM products were single-turn; today's products are agentic, where a model plans, calls tools, observes results, and loops until a task is done. This has become an engineering discipline of its own: you're managing state, cost, latency, termination, and failure modes for a component that is fundamentally non-deterministic.

In a nutshell, classical software engineering involves building deterministic software; AI engineering means designing reliable systems around an unreliable AI core, which means understanding guardrails, tool calling, retrieval, and evals.

What does an AI engineer do?

Day to day, AI engineers build cutting-edge applications that automate complex workflows, deliver more intelligent user experiences, and integrate AI into existing products and organizations. Typical projects include:

• Agentic chat and workflow products. Multi-turn assistants that call tools, query internal systems, and take actions — the AI customer support agent, the coding assistant, the research copilot.
• Retrieval-augmented generation (RAG) and search. Ingesting a company's documents, embedding and indexing them, and building the retrieval layer that lets a model answer questions about data it was never trained on.
• Evaluation and quality systems. The test suites of AI behavior: curated datasets, LLM-as-judge pipelines, human calibration loops, and CI gates that keep a prompt change from silently breaking production.
• Model serving and inference infrastructure. Routing requests to models, streaming tokens, managing GPU memory and batching, and controlling the cost curve as usage scales.
• Integrations and tooling. Exposing internal systems to agents safely, increasingly via standards like MCP, with the auth, latency, and safety questions that come with giving a model the ability to act.

What's the difference between ML engineers and AI engineers?

The key difference: ML engineers work at the model layer; AI engineers work at the application and platform layer.

ML engineers train and deploy the underlying models. Historically that meant classical techniques — regression, decision trees, recommendation models; today it increasingly means training, fine-tuning, and optimizing large transformer models for GPUs, creating pre-training data pipelines, and experimenting with new model architectures.

Applied AI engineers consume the outputs of those models at inference time and build products on top, then scale those products to potentially millions of users. This involves prompts and context management, orchestration and tool calling, retrieval, caching, evals, and the ordinary-but-essential distributed systems that surround all of it.

There's an overlap in the middle with post-training evaluation, fine-tuning, and embeddings. Plenty of engineers move between the two. In the interview, ML engineering loops focus more on modeling and math, while AI engineering loops go deeper on system design, product judgment, and the ability to ship. If your background is full-stack or backend engineering, AI engineering is the more natural on-ramp.

What the interview loop looks like

A full AI engineer loop at a top company or AI lab typically draws from five buckets:

1. System design, with agentic systems. Classic architecture questions still anchor the loop ("design Instagram"), but the distinctive rounds now ask you to design AI-native systems: an AI support agent, a RAG pipeline over enterprise documents, an LLM serving layer. The good news: these decompose into the same patterns as classic questions, plus a small set of new ones.
2. GenAI concepts. Conversational depth checks: what's a context window and what does it cost, why is long-context serving a memory problem, when would you fine-tune versus retrieve, how would you evaluate this feature? These show up as standalone rounds and as follow-up probes inside every other round.
3. Practical coding. Realistic project rounds: build from a spec, extend an unfamiliar codebase, or debug a misbehaving system — often with AI coding agents allowed or expected.
4. Classic coding. Expect to still be asked questions involving data structures and algorithms, especially at big tech.
5. Behavioral. How you've operated: ambiguity, conflict, shipping under pressure. In AI roles, expect questions about working with fast-changing technology and making judgment calls without established best practices.

What's in this track

Each area above has a full course, and each course pairs written lessons with videos, practice problems, and mock interviews:

• Overview (this course): the role, the market, and the process at top companies.
• System Design Interviews with AI: a pattern-based approach — ten architectural patterns that cover nearly every prompt, with dedicated modules on agentic architectures and model serving.
• GenAI Interviews and Concepts: LLM foundations (tokenizers, context windows, attention), embeddings and RAG, agents and MCP, prompt engineering, and evals.
• Practical Coding and Project Interviews: multi-file project rounds, take-homes, and debugging, organized along a read-to-write spectrum, with downloadable codebases modeled on real interviews.
• Coding Interviews: classic DS&A preparation.
• Behavioral Interviews: story banks, level calibration, and company-specific formats.

If you're starting from a software engineering background, follow the course in that order. If you already know your target company's loop, jump to the part of the course that maps to your upcoming rounds. Either way, the system design patterns and the GenAI concept lessons are important fundamentals to cover.