CS Fundamentals for AI Engineers — Overview

Someone who understands Transformers deeply but doesn't understand what actually happens between "I called model.generate()" and "a token appeared" isn't yet an AI systems engineer — they're someone who understands one layer of a much taller stack. Click a layer below to see the specific kind of real-world failure it explains:

CPU

↓

RAM

↓

PCIe

↓

GPU

↓

VRAM

↓

network

↓

distributed workers

Why a training job is slow even though "the GPU is only at 40% utilization" -- the CPU-bound data pipeline feeding it is the actual bottleneck.

None of that is ML theory — it's computer science, and this section exists because most ML curricula skip it entirely and most AI engineering jobs don't.

What's in this section

Operating Systems & Concurrency — processes, threads, concurrency vs. parallelism, memory, CPU caches, virtual memory. The layer that explains why your dataloader is slow and why two GPU processes can't cleanly share memory.
Networking & Distributed Systems — HTTP, TCP/IP, DNS, load balancing, the CAP theorem, consensus, and why distributed training and distributed serving are genuinely hard problems, not just "more machines."
APIs, HTTP & Communication Patterns — REST, auth (API keys, OAuth 2.0, JWTs), the five patterns that cover every way two systems stay in sync (polling, webhooks, SSE, WebSockets), and building one in FastAPI. The connective tissue behind every LLM API call, MCP server, and model-serving endpoint.
Linux, Git & Developer Tooling — the shell, the filesystem, and Git internals (not just commands) — the daily-driver tools every AI engineering job assumes fluency in.
Software Engineering Practice — design patterns, testing philosophy, profiling, and debugging, applied to the specific ways ML code breaks that ordinary software doesn't.

Why this comes first

Every later section in this site assumes you can read a stack trace, reason about why a process hung, and understand what "the network call timed out" actually means at the socket level. This section is the one prerequisite layer beneath Mathematics for AI — not because you need it to understand an equation, but because you need it to understand why that equation runs the way it does on real hardware.

See the roadmap for the full ordered path through this section.