Vibe Engines Learn
Vibe Engines Learn
Build-along course · RAG

Build a real AI app,
not another demo.

Most tutorials stop at a 10-line script that impresses in a screenshot and breaks in production. This course keeps going — eleven short episodes that take that toy and turn it into a retrieval-augmented app you'd actually ship. Every episode is embedded right here, with a written walkthrough and links to the deeper reference on each idea.

11 episodes ~2 hours Hands-on / build-along Free · no sign-up
Your progress0 / 11 complete

What you'll build

A production-minded RAG application: it ingests your documents, chunks and embeds them, retrieves the right passages with hybrid search and reranking, answers only from what it found with citations, becomes an agent that can call tools, is tested with real evals, runs fast with caching and streaming, is hardened against prompt injection, and is deployed with monitoring. You finish with the whole system — and the mental model of why each piece is there.

RAGChunkingEmbeddingsVector searchRerankingGrounding + citationsAgents / toolsEvalsCaching + streamingPrompt-injection defenseDeploy + monitoring
The curriculum

Eleven episodes, in order

01
Episode 1 · ~10 min

The Naive Version

The whole app in ten lines: stuff a document into the prompt and ask. It works on one page — and falls apart the moment your data is bigger than the context window. That failure is the reason everything else in this course exists.

What you'll learn
  • Why "just put it in the prompt" doesn't scale
  • The context-window wall you hit first
  • What retrieval-augmented generation (RAG) actually means
Go deeper
The RAG Handbook
02
Episode 2 · ~12 min

Chunking Makes or Breaks It

You can't embed a whole book as one vector. How you cut documents into chunks quietly decides whether retrieval ever finds the right passage — too big and the answer drowns in noise, too small and it loses the context that made it meaningful.

What you'll learn
  • Fixed-size vs semantic chunking
  • Why overlap matters
  • Why chunking is the highest-leverage knob in RAG
Go deeper
RAG Chunking playground The RAG Handbook
03
Episode 3 · ~13 min

Embeddings & Vector Stores

Turn every chunk into an embedding — a point in meaning-space — and store it so you can instantly find the ones nearest to a question. This is what makes "search by meaning, not keywords" possible.

What you'll learn
  • What an embedding really is
  • Nearest-neighbor search over vectors
  • Choosing and populating a vector store
Go deeper
The Vector Databases Handbook
04
Episode 4 · ~14 min

Hybrid Search + Reranking

Pure vector search misses exact terms — names, error codes, IDs. Combine it with keyword search, then let a reranker read the top candidates and reorder them. This is the single biggest quality jump in the whole course.

What you'll learn
  • Why dense retrieval quietly misses
  • Hybrid retrieval (BM25 + vectors)
  • Two-stage retrieval with a cross-encoder reranker
Go deeper
The Vector Databases Handbook The RAG Handbook
05
Episode 5 · ~12 min

Grounded RAG: Force Citations

Retrieval alone isn't enough — the model still makes things up. Force it to answer only from the retrieved passages, cite them inline, and say "I don't know" when the context doesn't contain the answer.

What you'll learn
  • Grounding the prompt in retrieved context
  • Inline citations users can trust
  • Honest refusal + measuring faithfulness
Go deeper
The RAG Handbook
06
Episode 6 · ~14 min

Turn It Into an Agent

Give the app tools and a loop — retrieve, decide, act, observe, repeat — and it becomes an agent that can look things up, call functions, and chain steps instead of answering in a single shot.

What you'll learn
  • The think → act → observe loop
  • Tool / function calling
  • When an agent beats one model call (and when it doesn't)
Go deeper
The Agent Patterns Handbook The MCP Handbook
07
Episode 7 · ~13 min

Testing & Evals (Stop Shipping on Vibes)

You can't ship on vibes. Build a golden set of question/answer pairs and use an LLM-as-judge to score retrieval and generation — so you catch a regression before your users do.

What you'll learn
  • Building a golden set
  • Retrieval metrics vs answer metrics
  • LLM-as-judge + regression-testing a RAG app
Go deeper
The LLM Evals Handbook The Agent Evals Handbook LLM-as-Judge builder
08
Episode 8 · ~12 min

Cache + Streaming (Why It’s Slow)

Your AI app is slow — and it usually isn't the model. Cache embeddings and answers, stream tokens as they generate, and the same app feels instant without touching quality.

What you'll learn
  • Where the latency actually goes
  • Caching layers that pay off
  • Token streaming + the latency-vs-cost dial
Go deeper
The LLM Serving Handbook Context-cost planner
09
Episode 9 · ~13 min

Prompt Injection & Security

A stranger's document can hijack your bot — "ignore your instructions and…" hidden inside retrieved text. Learn exactly how the attack works, then the defenses that actually hold.

What you'll learn
  • How injection rides in through retrieved content
  • Trust boundaries + input/output guarding
  • Least-privilege tools
Go deeper
Prompt-Injection Playground (lab) The MCP Handbook
10
Episode 10 · ~13 min

Deploy, State & Monitoring

Shipping is 90% boring plumbing — state, secrets, rate limits, logging, and watching the thing drift in production. It's the part tutorials skip and production demands.

What you'll learn
  • Statelessness + session state
  • Secrets + config the right way
  • Observability for LLM apps + catching silent drift
Go deeper
The Observability Handbook
11
Episode 11 · ~10 min

Finale — You’re an AI Engineer

Step back and see the whole system you built — and why assembling these pieces well, not prompt-wizardry, is what the job actually is. Then: where to go next.

What you'll learn
  • The full architecture, end to end
  • What separates a demo from a product
  • Your next steps as an AI engineer
Go deeper
The AI Engineer Roadmap Senior AI Engineer Handbook

Finished the build?

You've assembled a real retrieval-augmented app end to end. The natural next step is the wider craft — the roadmap that frames where this sits, and the interview handbook that pressure-tests it.

The AI Engineer Roadmap → Senior AI Engineer Handbook