A few weeks ago I watched someone I know get a gig building an AI model for an insurance company. The brief: simulate a car crash, reconstruct the physics, figure out who was at fault. She used physics simulations, machine learning, computer vision. Things I'd heard of but couldn't build. I sat with that feeling for a while.

So I made a decision. I'm going to learn machine learning — properly, not just prompt engineering and vibes. And I'm going to document every single step here so that in six months, there's a record of exactly where I started.

Today was day one.

What I built

A Python script that fetches real-time weather data from any city in the world and prints it in plain English.

import requests 
import sys

def get_weather(city): 
    url = f"https://wttr.in/{city}?format=j1" 
    response = requests.get(url) 
    data = response.json()

    temp = data["current_condition"][0]["temp_C"]
    feels_like = data["current_condition"][0]["FeelsLikeC"]
    description = data["current_condition"][0]["weatherDesc"][0]["value"]

    return {
        "city": city,
        "description": description,
        "temp": temp,
        "feels_like": feels_like
    }

if __name__ == "__main__":
    city = sys.argv[1] if len(sys.argv) > 1 else "Nairobi"
    result = get_weather(city)
    print(f"{result['city']} right now: {result['description']}")
    print(f"Temperature: {result['temp']}°C (feels like {result['feels_like']}°C)")

I ran it for Nairobi, then Tokyo, then London. It just worked.

What I actually learned

I come from Java so Python syntax wasn't the hard part. What was new: how Python handles libraries (import requests instead of Maven dependencies), how JSON from an API becomes a Python dictionary automatically via response.json(), and why every project gets its own virtual environment instead of installing packages globally.

The line that made me think hardest was this one:

temp = data["current_condition"][0]["temp_C"]

That's three operations chained: get a list from a dict, grab the first item, then get a value from that dict. Once you see it, you can't unsee it. That's how you navigate any API response.

The setup

Mac M4, Python managed by uv (think Gradle but for Python versions and packages), VS Code with the Python and Jupyter extensions. Setup took under 20 minutes.

What's next

Tomorrow I extend this into a morning brief bot — weather, news, calendar — that runs automatically every day. By end of September, I want to have a solid enough Python base that I can start touching real data science libraries.

The destination is that insurance model. I'm not there yet. But today I made the first commit.

Nairobi, September 2026. Let's go.