Notebook vs Production Approach – Ai Engineering Core

🔁 Iteration is Essential

When AI output doesn't match expectations:

❌ Don't immediately change the code
✅ First improve the prompt

📤 Generate Output
       ↓
🔍 Evaluate Result
       ↓
✏️ Adjust Prompt
       ↓
🔄 Generate Again
       ↓
🔁 Repeat until satisfied

🎯 Prompt iteration is one of the most important skills when building AI applications!

🧪 Notebook vs Production Development

📓 Notebook Style — "Data Scientist Mode" 🔬

In notebooks, we act as experimenters, not engineers. Variables are defined globally and freely overridden as we iterate:

# Define it...
SYSTEM_PROMPT = "You are a helpful assistant."

# Didn't like it? Just overwrite it! 🔄
SYSTEM_PROMPT = "You are a funny assistant."

# Still not right? Override again! 🔄
SYSTEM_PROMPT = "You are a pirate assistant. Arrr! 🏴‍☠️"

🧠 As a data scientist, you're exploring and experimenting — the goal is to find what works, not to write clean code. Global variables are a shortcut that makes iteration fast and frictionless.

✅ Great for...

⚡ Fast experimentation

🧪 Rapid prompt testing

✏️ Quick tweaks without restructuring

🏭 Production Style — "Software Engineer Mode" 👷

As a software developer, overriding global variables is a red flag 🚩. Instead, everything is structured, controlled, and reusable:

# ✅ Prompts live inside functions/classes — never floating globally
class BrochureGenerator:
    def __init__(self, system_prompt: str):
        self.system_prompt = system_prompt  # injected, not hardcoded

    def generate(self, company_name: str, url: str):
        ...

Pattern	Purpose
🧱 Classes	Organize logic cleanly
🔧 Functions	Reusable, testable code
💉 Dependency Injection	Pass prompts in — don't hardcode them
⚙️ Config Management	Manage prompts & settings from one place

📡 Streaming Protocols

Generated tokens reach the user instantly using technologies like:

Technology	Use Case
📨 Server-Sent Events (SSE)	Most common for AI streaming
🌊 Streaming HTTP	Simple, efficient token delivery
🔌 WebSockets	Real-time, two-way applications

⚡ These technologies ensure tokens appear on screen immediately as they're generated — no waiting for the full response!