Levi Lee
EN 中文

May 28, 2026

Setting Up Your AI Dev Environment from Scratch

This was the first class I taught for NTU AI Club 114-2, built together with datafox.tw — a “survival kit” that gets someone who has never written a line of code to a working environment and a first running program in one evening.

I graduated in political science and only switched into software later, so I know exactly how off-putting “just getting the environment set up” can be for a beginner. That’s why this is written in painful detail — just follow along.

Why Python?

Pros

  • Clean, intuitive syntax — close to English, beginner-friendly, fast to pick up.
  • A complete ecosystem with easy package management; AI/ML, data analysis, and Web all have mature libraries.
  • A huge community — rich resources on Stack Overflow and GitHub.
  • Industry standard — Google, Meta, and OpenAI all use it heavily.

Things to note

  • Slower (it’s interpreted — 10–100× slower than C/C++).
  • The GIL (Global Interpreter Lock) limits multithreading.
  • Loose typing can cause bugs in large projects.

But in AI/data work almost everything is Python, so none of this gets in the way of learning. Bottom line: for learning AI, Python is the best starting point right now.

What are an IDE and a Terminal?

Get these two terms straight first so the rest makes sense:

  • IDE (Integrated Development Environment) — where you write code. It gives you syntax highlighting, autocomplete, a debugger, file management, Git integration, and usually a built-in terminal.
  • Terminal — where you talk to the computer with text commands: run Python programs, install packages (pip install), work with the file system. It’s “Terminal” on macOS and “CMD / PowerShell” on Windows.

Simple version: the IDE is where you write code, the terminal is where you run commands, and the terminal lives inside the IDE.

Recommended IDEs

  • VS Code — free, classic, lightweight, tons of extensions.
  • Antigravity — Google’s AI IDE (late 2025). It’s built on VS Code, so if you can use VS Code you can use it; it just adds a built-in Gemini AI Agent that can write code and run tests for you.

Installing Python

macOS

  1. Download the installer from python.org/downloads; or
  2. If you have Homebrew, just run brew install python3.

Windows

Download the installer from python.org/downloads. Make sure you check “Add Python to PATH” during installation, otherwise the terminal won’t find python.

Verify it worked — open a terminal and run:

python3 --version    # macOS
python --version     # Windows

If you see a version number (e.g. Python 3.12.x), you’re good.

VS Code / Antigravity extensions

  • Must-have: Python, Jupyter.
  • Optional: WakaTime (tracks your coding time — surprisingly motivating), Cline (a free AI coding assistant, a good fallback when your Gemini quota runs out), Rainbow CSV (read CSV files more clearly).

Package management: start with uv

pip is Python’s package manager — think “App Store for Python”. Common commands:

pip install requests              # install a package
pip install requests==2.31.0      # pin a version
pip list                          # list installed packages
pip freeze > requirements.txt     # export the list
pip install -r requirements.txt   # install everything in the list

That said, I’d suggest beginners start with uv. Built by Astral (the team behind Ruff), it replaces pip + venv + pyenv in a single tool, installs packages 10–100× faster than pip, and manages virtual environments for you — saving you a pile of fiddly setup.

# Install uv
# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows (PowerShell)
irm https://astral.sh/uv/install.ps1 | iex

# Create a project & add packages
uv init my-ai-project       # new project, auto-generates pyproject.toml
cd my-ai-project
uv add requests             # add a package, way faster than pip install
uv add google-genai         # add the Gemini SDK
uv run python main.py       # runs with the correct virtual env automatically

What are an API and an API Key?

Building anything with AI means using an API: your program “calls” someone else’s service (like Gemini) through it. And your API Key is your identity — anyone who gets it can call the service as you, which means spending your money.

A real case: someone pushed an API Key to GitHub and got billed US$55,000.

So whenever, wherever: never leak any of your API keys, and never put them in your code or push them to a public repo.

Getting a free Gemini API Key

Free, no credit card needed — a Google account is all it takes:

  1. Open Google AI Studio and sign in with your Google account.
  2. Accept the Terms of Service on first login.
  3. Click “Get API Key” on the left (or go straight to aistudio.google.com/apikey).
  4. Choose “Create API key in new project” — Google creates the Cloud project for you.
  5. Copy it and save it somewhere safe immediately. The key is a long string starting with AIza.

For learning and small projects, the Free Tier is more than enough; if you go over the limit you’ll get a 429 (Rate Limit) error — just wait a moment and try again.

The most important part: key safety

Never hardcode your key:

api_key = "AIza..."   # ← never do this

The right way is to use an environment variable, keep it in .env, and add .env to .gitignore so it’s never uploaded by git:

# in your terminal or shell config
export GEMINI_API_KEY="your-key"
.env

Today’s tasks

  1. Install Python and an IDE (VS Code or Antigravity).
  2. Get Hello, World! to run.
  3. Watch one recommended Python tutorial.
  4. (Optional) Get a Gemini API Key and use it to build a single-turn chatbot to test it out.

Once the environment’s up, the next step is syntax. I put that in a separate post: Python 101: The Syntax You Actually Need Before Your First AI Project.

Switching fields is genuinely hard — but once you clear “get the environment working”, you’ve already taken the toughest step. Questions welcome by email; I’m happy to chat.

Writing