GenAI Architecture Guide for FDE & Applied AI Interviews

Learn how to design and explain production GenAI systems—from model APIs and RAG to agents, evaluation, security, and operations.

This guide is written for Forward Deployed Engineer (FDE), Applied AI Engineer, and AI Engineer candidates who need to reason about customer-facing AI systems, not memorize a fashionable framework.

Start the complete FDE GenAI course →

What makes this guide different

Architecture decisions, not framework tutorials. Learn when to use a direct model call, RAG, a deterministic workflow, or an agent.
Production trade-offs, not toy chatbots. Treat permissions, evaluation, latency, cost, observability, and recovery as part of the design.
Structured for system-design interviews. Build the vocabulary and decision process expected in FDE and Applied AI interviews.
Connected to deliberate practice. Continue into guided architecture scenarios, coding exercises, and an AI mock interviewer when you are ready to rehearse.

Start here

Architecture pattern selection cheat sheet

Start with the simplest pattern that satisfies the requirement. Add autonomy only when the system genuinely needs to choose its path at runtime.

Requirement	Start with	Add only when needed
Generate, classify, extract, or transform bounded input	Direct model call	Structured output and deterministic validation
Maintain a conversational interface	Chatbot	Explicit history selection and state
Answer from private, current, or citable knowledge	RAG	Hybrid retrieval, reranking, and access-aware filtering
Execute a known multi-step business process	Deterministic workflow	Model steps inside application-controlled transitions
Choose tools or next steps dynamically	Agentic workflow	Budgets, stop conditions, scoped permissions, and recovery
Perform an irreversible or high-risk action	Human approval	Audit evidence and explicit authorization before execution

Before choosing a pattern, ask:

Does the model already have the required information?
Are the execution steps known in advance?
Must the system take an external action?
What is the cost of a plausible but wrong result?
How will we evaluate, trace, recover, and stop it?

Course contents

Chapter	What it covers
01 · Foundations of GenAI Systems	Six lessons, from traditional software to the new role of the engineer
02 · Building Blocks of GenAI Systems	Models, context, retrieval, tools, MCP, state, memory, and permissions
03 · GenAI Architecture Patterns	Direct LLM, RAG, workflows, agents, skills, interoperability, and human review
04 · Reliable AI Systems	Evaluation, security, guardrails, observability, governance, and readiness
05 · Production-Grade GenAI Engineering	Specifications, production architecture, deployment, data, CI/CD, scale, and operations

Every website subsection is stored as its own numbered Markdown file with the same lesson title.

Open guide vs. complete handbook

This repository explains the core mental models and production principles. The complete FDE Handbook adds deeper course chapters, guided walkthroughs, runnable coding exercises, closed-book practice, progress tracking, and an AI mock interviewer.

Locked lessons use the same Continue the course or Member chapter boundary shown in the Handbook.

Continue with practice

The previous question-bank README is preserved in archive/question-bank.md.

License

Course content is available under CC BY-NC 4.0. You may share and adapt it with attribution for non-commercial use. The synchronization script is available under the MIT License.

Built for engineers preparing for Forward Deployed Engineer, Applied AI Engineer, Solutions Engineer, Customer Engineer, and Field Engineer roles.

genai-architecture-guide