Skip to content
CodeStoa

AI Engineering

Learn AI engineering one topic at a time: how LLMs work, prompts, RAG, agents, evals and production, from beginner to interview-ready.

18 modules14 lessons75 min read

Start here

Follow the AI Engineer roadmap, beginner to expert

start here

Beginner 4:23 video

What Does an AI Engineer Do? The Three Jobs Behind AI Apps

Module 01

Start Here

What the job is, how it differs from ML and data science, and where LLMs fit in AI.

Start module
  1. 01

    What Does an AI Engineer Do? The Three Jobs Behind AI Apps

    An AI engineer takes a ready-made AI model and builds a real product around it. One small lunch-box business shows the three jobs that turn a guessing chat assistant into one you can trust.

    Beginner4 min readNov 9, 2026 Video
  2. 02AI Engineer vs ML Engineer vs Data ScientistSoon
  3. 03AI, Machine Learning, Deep Learning and LLMsSoon

Module 02

How LLMs Work

Tokens, next-token prediction, the context window, training, sampling settings and why models make things up.

Start module
  1. 01

    What Is a Large Language Model (LLM)? Why It Can Be Confidently Wrong

    A large language model writes a reply by guessing what usually comes next, one piece at a time. That makes it brilliant at writing and explaining, and confidently wrong about facts it never read.

    Beginner6 min readOct 28, 2026 Video
  2. 02

    What Are Tokens in AI? Why Some Models Miscount the R's

    An LLM never reads letters or whole words. It reads tokens: pieces of text, each swapped for a number. Here's how that works, and why it decides limits, prices and spelling mistakes.

    Beginner5 min readOct 29, 2026 Video
  3. 03

    Next-Token Prediction: How an LLM Writes One Token at a Time

    An AI chat never has a finished answer waiting. It scores every possible next token, picks one, adds it to the text and runs again, until it picks a stop token.

    Beginner5 min readOct 31, 2026 Video
  4. 04

    Context Window Explained: Why an AI Model Forgets

    An AI model can only see a fixed amount of text at once: its context window. When your text doesn't fit, part of it is cut, and the model seems to forget.

    Beginner3 min readOct 26, 2026 Video
  5. 05How an LLM Is Trained: Pre-training to RLHFSoon
  6. 06Temperature, Top-k and Top-pSoon
  7. 07Why LLMs HallucinateSoon

Module 03

Choosing a Model

Open vs closed, the main model families, reasoning models, and trading quality against speed and cost.

Start module
  1. 01Open vs Closed ModelsSoon
  2. 02Model Families: GPT, Claude, Gemini, LlamaSoon
  3. 03Reasoning Models vs Chat ModelsSoon
  4. 04

    What Is Jev? A Decision Model That Picks Instead of Writing

    An LLM writes an answer word by word. A decision model like Jev picks one option from your list and says how sure it is. Here's how that works, what it costs, and where it breaks.

    Intermediate6 min readOct 24, 2026 Video
  5. 05Choosing a Model: Quality, Speed and CostSoon

Module 04

Prompt Engineering

Writing prompts that work every time: roles, examples, step-by-step reasoning and output you can parse.

  1. 01System, User and Assistant MessagesSoon
  2. 02Zero-Shot vs Few-Shot PromptingSoon
  3. 03Chain of Thought PromptingSoon
  4. 04Structured Output: Getting JSON BackSoon
  5. 05Writing Prompts That Don't BreakSoon

Module 05

Calling Models From Code

Your first API call, streaming, what each call costs, prompt caching, and handling failures.

  1. 01Your First LLM API CallSoon
  2. 02Streaming ResponsesSoon
  3. 03Tokens, Cost and Rate LimitsSoon
  4. 04Prompt CachingSoon
  5. 05Retries, Timeouts and Fallback ModelsSoon

Module 06

Embeddings and Vector Search

Turning text into numbers that capture meaning, measuring how close two texts are, and searching millions fast.

  1. 01What Are Embeddings?Soon
  2. 02Cosine SimilaritySoon
  3. 03Semantic Search vs Keyword SearchSoon
  4. 04Vector DatabasesSoon
  5. 05How Vector Indexes Work: HNSWSoon

Module 07

RAG

Retrieval-augmented generation: answering from your documents, and fixing it when the answer is wrong.

  1. 01What Is RAG?Soon
  2. 02Chunking StrategiesSoon
  3. 03Retrieval: Top-k and Metadata FiltersSoon
  4. 04Hybrid Search: Keywords Plus VectorsSoon
  5. 05RerankingSoon
  6. 06Why RAG Fails and How to Fix ItSoon
  7. 07RAG vs Fine-Tuning vs Long ContextSoon

Module 08

Evaluating LLM Apps

Measuring quality instead of guessing: failure points, criteria, methods, test sets and regression checks.

Start module
  1. 01

    LLM Evals: Why "It Looks Good" Isn't Enough

    LLM evals are systematic, repeatable tests that judge an LLM app against clear criteria. Learn why spot checks fail, how evals differ from classic tests, and the workflow to run them.

    Beginner4 min readOct 19, 2026 Video
  2. 02

    Failure Points in LLM Applications: Where RAG Apps Break

    A RAG chatbot can fail in its retriever, its generator, the flow between them or the app as a whole. Follow one customer's question to see why checking only the final answer can't tell you which part broke.

    Beginner6 min readOct 22, 2026 Video
  3. 03

    LLM Evaluation Criteria: Quality, Safety and Operational Metrics

    Every LLM eval criterion answers one of three questions: is the answer any good, could it cause harm, and is it fast, cheap and reliable enough to run? Follow one support bot through each of them.

    Beginner7 min readOct 22, 2026 Video
  4. 04

    LLM Eval Methods: Code Checks, Human Review and LLM-as-Judge

    An eval method is the procedure that turns one output into a verdict. See code, a judge model and a person each judge the same five customer emails, and learn when each one is the right tool.

    Beginner6 min readOct 23, 2026 Video
  5. 05Building a Golden Test SetSoon
  6. 06Evaluating RAG: Faithfulness and RecallSoon
  7. 07Regression Testing PromptsSoon

Module 09

Tool Calling and Agents

The loop that turns a model into an agent, how it picks tools, how it remembers, and the limits that keep it safe.

Start module
  1. 01

    The Agent Loop: How an LLM Plans, Acts and Knows When to Stop

    An agent is a model inside a loop: it reads the goal and the history, picks the next tool, sees the result, and repeats. Loop engineering is designing that loop on purpose, including how it stops.

    Beginner6 min readOct 24, 2026 Video
  2. 02

    Tool Calling: How an Agent Picks and Uses a Tool

    The model never runs a tool. It writes a call, a tool name plus arguments, and your code checks it, runs it and hands back the result. Here is one round trip, start to finish, at a small cinema.

    Beginner6 min readOct 24, 2026 Video
  3. 03ReAct: Reason, Then ActSoon
  4. 04Agent Memory: Short-Term and Long-TermSoon
  5. 05

    AI Agent Guardrails: Limits That Keep an Agent Safe

    Asking the model nicely in the prompt is not a guardrail. A guardrail is a check your code runs around the model, and it can say no whatever the message says.

    Beginner6 min readOct 25, 2026 Video

Module 10

MCP

The Model Context Protocol: one standard way to plug any tool into any agent.

Start module
  1. 01

    What Is MCP? How an AI Assistant Plugs Into Your Tools

    MCP, the Model Context Protocol, is one shared plug between AI apps and tools. Follow Alex asking their assistant to book an hour with Bob, and see 12 custom connections become 7 pieces.

    Beginner5 min readOct 25, 2026 Video
  2. 02Building an MCP ServerSoon
  3. 03MCP Transports: Local vs RemoteSoon

Module 11

Multi-Agent Systems

When a fixed workflow beats an agent, how several agents split work, and how to judge a whole run.

  1. 01Workflows vs Agents: When Not to Use an AgentSoon
  2. 02Multi-Agent Patterns: Orchestrator and WorkersSoon
  3. 03How Coding Agents WorkSoon
  4. 04Evaluating an AI AgentSoon

Module 12

Context Engineering

Deciding what the model sees on every call, and the ways a crowded context makes it worse.

  1. 01Prompt Engineering vs Context EngineeringSoon
  2. 02Context Sources: What Goes Into the WindowSoon
  3. 03Context CompactionSoon
  4. 04Lost in the Middle: Long-Context LimitsSoon
  5. 05How Context FailsSoon

Module 13

Safety and Security

Prompt injection, guardrails on input and output, private data, bias, and attacking your own app first.

  1. 01Prompt InjectionSoon
  2. 02Input and Output GuardrailsSoon
  3. 03PII and Private Data in LLM AppsSoon
  4. 04Bias and FairnessSoon
  5. 05Red-Teaming an LLM AppSoon

Module 14

Observability and Cost

Seeing what your app does in production and keeping the bill under control.

  1. 01Tracing an LLM AppSoon
  2. 02Monitoring Cost and LatencySoon
  3. 03Monitoring Quality in ProductionSoon
  4. 04Semantic Caching and Model RoutingSoon

Module 15

Serving Models

Running models yourself: locally, at scale with batching and the KV cache, smaller with quantization.

  1. 01Running an LLM on Your LaptopSoon
  2. 02How Inference Serving Works: The KV CacheSoon
  3. 03Batching and ThroughputSoon
  4. 04QuantizationSoon
  5. 05Self-Hosting vs API: The Cost MathSoon

Module 16

Fine-Tuning

When to train a model on your data, how LoRA makes it cheap, and how to check it helped.

  1. 01When to Fine-Tune a ModelSoon
  2. 02LoRA: Fine-Tuning on a BudgetSoon
  3. 03Preparing Training DataSoon
  4. 04Distillation: A Small Model Learns From a Big OneSoon

Module 17

Multimodal AI

Models that see, hear and speak: images in and out, speech and real-time voice.

  1. 01Vision Models: How an LLM Reads an ImageSoon
  2. 02How Image Generation WorksSoon
  3. 03Speech to Text and Text to SpeechSoon
  4. 04Building a Voice AgentSoon

Module 18

AI System Design Interviews

A step-by-step method, then five classic AI systems designed from requirements to evals.

  1. 01How to Approach an AI System Design InterviewSoon
  2. 02Design a Chat-With-Your-Docs AppSoon
  3. 03Design a Customer Support AgentSoon
  4. 04Design a Semantic Search EngineSoon
  5. 05Design an AI Code Review BotSoon
  6. 06Design an LLM GatewaySoon