Learn AI · ShipAI

Browse concepts

Hello Interview–style topic atlas — Concepts, Tools, and Data & inference. Article-first; each page links into the guided lab when one exists.

Concepts — Foundations → deeper theory

Core Concepts

Foundations every AI engineer needs · 16 articles

Advanced Concepts

Harder theory once foundations click · 8 articles

Tools — Everyday stacks → production tooling

Key Tech

Stacks you will actually use · 8 articles

Advanced Key Tech

Scale and production tooling · 6 articles

Data & inference — Retrieval data plane + serving

Data & Databases for AI

Vectors, caches, retrieval data plane · 6 articles

Inference

Serve fast, cheap, and reliably · 6 articles

Guided path (build projects)

Ordered modules with milestones — the project spine. Start at Onboarding; skim early theory if you already ship software. Maps to browse concepts on the outline.

Foundations

Module	Lessons	Outcome
Onboarding Set up the portfolio monorepo, tooling, API hygiene, and the tests/eval/README rubric so the lab is ready.	4	Lab ready.
What happened in AI (till now) Map symbolic AI → classical ML → deep learning → transformers → generative systems, then place yourself with a gap list.	7	Field-map write-up (blog-style lesson output).
ML/DL literacy Practical literacy: splits/leakage, loss and overfitting, embeddings intuition, autograd, and a real PyTorch training loop.	5	Training-loop repo section.
Build an LLM from scratch Karpathy-depth, course-paced: tokenize, predict, attend, and train your own mini-LLM — then state honest limits vs frontier systems.	8	Your own mini-LLM in the portfolio (train script, samples, README).

Generative AI practice

Module	Lessons	Outcome
Talk to models in the real world Chat APIs, structured output, prompt versioning, multimodal I/O, and code-assist loops with eval discipline.	5	Versioned prompts + eval set.
RAG Embeddings, chunking, vector DB, hybrid search, rerank/rewrite, and citation/failure drills for grounded generation.	6	Domain RAG with grounded eval.
Build & serve your SLM Decide when FT wins, curate data, LoRA/QLoRA, eval vs baselines, serve locally, and measure quantization tradeoffs.	6	Your SLM — fine-tuned + served + beating prompt-only baseline on a narrow task.

Agents & agentic systems

Module	Lessons	Outcome
Build real AI agents Hand-rolled ReAct, real tools, planning, memory, HITL, trajectory debugging, then framework packaging.	7	Production-shaped single agent (≥4 tools, memory, HITL, evals).
Skills, MCP, context engineering Package skills, load them dynamically, connect MCP servers/clients, and budget context like an engineer.	6	Skill pack + MCP + host agent.
Agentic workflows & multi-agent Router/orchestrator patterns, handoffs, agentic RAG, durable jobs, failure isolation, and harness loops.	6	Multi-agent / agentic workflow with evals + failure test.

Production + how companies actually do it

Module	Lessons	Outcome
Evals, guardrails, safety Offline/online evals, judges, component vs E2E, red teams, guardrails, and canary gates on your best agent.	6	Eval + guardrails on your best agent.
Deploy, cost, latency, observability Streaming APIs, cost logs, caching, traces, resilient clients, and a runbook for your service.	6	Deployed agent/SLM service with cost + traces.
How real companies use AI Case labs: read public eng posts, extract architecture, rebuild a tiny analogue, compare tradeoffs across major companies.	8	Portfolio folder industry/ with lab reports + one mini-system mirroring a company pattern.
AI system design Synthesize build-vs-buy, UX of uncertainty, data flywheels, privacy, and a graded capstone proposal citing industry labs.	5	Capstone design doc citing ≥2 industry patterns from Industry case labs.

Frontier + capstone

Module	Lessons	Outcome
What’s coming + staying current Reasoning/test-time compute, multimodal agents, on-device models, synthetic data, safety, and a personal learning radar.	6	State of AI brief + one frontier mini-demo.
Capstone Course exit: ship a system with at least two pillars (owned model, agent, agentic workflow, industry pattern) plus evals and a write-up.	4	Repo, architecture, eval report, written post (optional video later).