Lesson 08 — Putting it together

From “I use AI in chat” to “I built something with AI.”
A practical road map for real beginners.

AI for Dummies Module 8 18 slides

Covers: cloud vs local decision · 4 starter projects · what RAG is · the 3-layer app model · Claude Code scaffolding · 4 golden rules

Decision #1 — Cloud or local?

Four quadrants: Privacy sensitivity (vertical) × Expected volume (horizontal).

The interactive wizard on the Theory page scores your situation across 5 dimensions and gives a personalised recommendation.

Cloud option — deep dive

Local option — deep dive

“Prototype cloud, scale local” — the smart arc

Start fast with cloud quality. Swap to local when cost or privacy become real concerns. One line of code changes.

Starter project #1 — PDF Summariser

Loop through a folder of PDFs, send each one to the model, write summaries to a text file. Perfect first project.

import anthropic, pathlib

client = anthropic.Anthropic()   # reads ANTHROPIC_API_KEY from env

for pdf_path in pathlib.Path("reports/").glob("*.pdf"):
    text = extract_text(pdf_path)            # PyMuPDF or pdfplumber
    response = client.messages.create(
        model="claude-haiku-4-5",
        max_tokens=300,
        messages=[{"role": "user", "content":
            f"Summarise in 3 bullet points:\n\n{text[:3000]}"}]
    )
    print(f"=== {pdf_path.name} ===")
    print(response.content[0].text)

Starter project #2 — Message Sorter

Classify incoming emails or support tickets into categories. The AI returns a label — you route accordingly.

{"category": "billing",
 "confidence": "high",
 "reason": "mentions invoice"}

Starter project #3 — Personal Q&A Bot

Point an AI at a folder of your own notes or docs. Ask it questions. It answers from your material.

Starter project #4 — Auto-Reply Drafter

Draft replies to common inbound questions in your tone. Human reviews & hits send. Speed up support by 3×.

What is RAG? — the core idea, simply

RAG flow — 5 stages (animated)

Watch the 5 stages appear in sequence. Each is explained below the diagram.

The animated RAG diagram on the Theory page has hover tooltips explaining each stage in detail.

“Chat with your own docs” — tools that handle RAG for you

You don't need to build the search engine. These tools do the whole RAG pipeline with a GUI.

The 3-layer app architecture

Every AI application, from a weekend script to a shipped product, is made of the same three layers.

Claude Code can scaffold this for you

Describe the app you want in plain English. The agent builds the boilerplate, wires the API calls, and sets up the folder structure. You steer and review.

Build a Python script that:
1. Reads every .txt file in ./emails/
2. Classifies each as billing,
   support, or spam
3. Outputs a CSV: filename, category,
   confidence
4. Uses Claude Haiku, reads ANTHROPIC_API_KEY
   from env

Golden Rules #1 & #2

Golden Rules #3 & #4

A first-weekend project plan

The “Meeting Notes → Action Items” project. Completable in 90 minutes with a free Claude account and a text editor.

Recap & next: Safety