Jupyter & Colab

The Science Lab Analogy

Remember school science labs? You had a notebook where you wrote the experiment, did it, noted the result, then tried the next experiment - all in the same notebook. Jupyter is EXACTLY that for coding!

Instead of writing a complete program, running it, and seeing one final output, Jupyter lets you run code in small chunks (cells), see results immediately, and iterate quickly. It's like having a conversation with Python.

Why AI Engineers Use Jupyter:

• Instant feedback - Run one cell, see the result, adjust. No need to re-run entire scripts.
• Mix code + text + images - Explain your thought process alongside code. Perfect for documentation.
• Data visualization - Charts, graphs, and tables render inline. See your data come alive.
• Experimentation - Try different prompts, parameters, models. Keep all experiments in one place.
• Sharing - Share .ipynb files with colleagues. They see your code, results, AND explanations.

Jupyter vs Regular Python Scripts:

Python script (.py) = Like writing a letter. You write the whole thing, then run it all at once.

Jupyter notebook (.ipynb) = Like a WhatsApp chat with Python. You send a message (code), get a reply (output), then continue the conversation.

What is Google Colab?

Google Colab is a FREE Jupyter Notebook that runs in the cloud on Google's servers. The killer feature? FREE GPU access! You can run machine learning models, fine-tune LLMs, and process data without buying expensive hardware.

Think of it like Google Docs but for code. No installation, runs in browser, auto-saves to Google Drive, and you can share with anyone via a link.

Colab Free vs Pro vs Pro+:

• Free - T4 GPU, 12GB RAM, disconnects after idle. Good for learning and small experiments.
• Pro ($10/month) - A100 GPU, 25GB RAM, longer sessions. Good for serious projects.
• Pro+ ($50/month) - Priority GPUs, background execution. For heavy AI work.

Colab Superpowers:

• Install any library with !pip install
• Mount Google Drive for persistent storage
• Use shell commands with ! prefix
• Free access to GPUs (T4, sometimes A100)
• Share notebooks like Google Docs
• Built-in forms for interactive inputs

Must-Know Keyboard Shortcuts:

Shift + Enter    = Run cell and move to next
Ctrl + Enter     = Run cell and stay
Alt + Enter      = Run cell and insert new below
A                = Insert cell above (command mode)
B                = Insert cell below (command mode)
DD               = Delete cell (command mode)
M                = Convert cell to Markdown
Y                = Convert cell to Code
Esc              = Enter command mode
Enter            = Enter edit mode
Tab              = Code completion
Shift + Tab      = Show function documentation

Cell Types:

• Code cells - Write and run Python code. Output appears below.
• Markdown cells - Write formatted text, headings, bullet points. Use for documentation.
• Raw cells - Plain text, no formatting. Rarely used.

AI Workflow in Jupyter:

1. 1. Install libraries and import modules
2. 2. Set up API keys and configurations
3. 3. Load and explore your data
4. 4. Experiment with prompts/models (iterate fast!)
5. 5. Analyze results with visualizations
6. 6. Document findings in Markdown cells
7. 7. Export clean code to .py for production

Mistake 1: Running Cells Out of Order

Notebooks let you run cells in any order. This creates hidden state bugs. Cell 5 might depend on Cell 3 which you modified but didn't re-run. Always do "Restart Kernel & Run All" before sharing.

Mistake 2: Putting API Keys in Notebooks

NEVER hardcode API keys in notebooks! Use environment variables or Colab's Secrets feature. Notebooks get shared - your API key could leak.

Mistake 3: Using Notebooks for Production

Notebooks are for experimentation, not production. Once your code works, extract it into proper .py files with functions, error handling, and tests.

Best Practices:

• Keep cells small and focused (one task per cell)
• Use Markdown cells to document your thought process
• Clear all outputs before committing to git
• Use nbstripout to auto-strip output on commit
• Name notebooks descriptively: "02_prompt_experiments_rag.ipynb"

Q1: When would you use Jupyter Notebooks vs Python scripts?

Answer: Notebooks for exploration, prototyping, data analysis, and documentation. Scripts for production code, automation, APIs, and anything that needs to run unattended. Many teams prototype in notebooks then convert to scripts.

Q2: How do you handle API keys in Google Colab?

Answer: Use Colab's built-in Secrets feature (key icon in left sidebar), or mount Google Drive and read from a .env file, or use environment variables. Never hardcode keys in notebook cells.

Q3: What is a Jupyter kernel?

Answer: The kernel is the computation engine that executes the code in cells. It maintains state (variables, imports) between cell executions. Restarting the kernel clears all state. You can have different kernels for different languages (Python, R, Julia).