And now we have 100 subs!!

When the channel was created back on 2023, it was private and i was the only sub Just as a telegram version of my journal and diary since i cannot have my journal book with me everytime. It was where i put everything that crossed my mind hence the name "rumbling".(most of the original content back the are deleted now😅). And few months ago i decide to make it a public. And look at where we are! And I am thankful for each and every of you. 🫶 I will also try to minimize the rumbling and make worthy of my audience ✨❤️

My friends and I have been thinking about a startup idea, but before diving into it blindly, we need to do some research and collect real feedback.

And here I am asking for a little help from my people.

Our target group is beauty professionals: hairstylists, makeup artists, nail technicians, henna artists, salon owners, trainers, etc.

So if you know anyone in this space, please share this form with them and help me get it in front of the right people. If you know someone who works in a beauty salon or training center, I’d especially appreciate it if you could forward it to them.

Also, if you’re part of the target group yourself, please fill it out! Just share it for me and help me reach the right people. 🤝

For any Questions DM me at : Jennah
Form: Link

Neverrrrrrr accept jokes about

your Family
your body
your partner
your dreams,
your trauma
your job
Your insecurities
or the way you dress.

Disrespect comes in the form of jokes

Have a beautiful day💐
#Fromyfyp
@Savvy_Society

So, as for starting this week, I created an SMS Spam Classifier project, which is a simple ML model that detects spam text messages using TF-IDF + Logistic Regression.This is basically a transition project before I move on to other linear models and eventually neural networks.
The classifier is actually much simpler than it initially seems.
After cleaning the data, what I really did was:
Text → TF-IDF numerical features → Logistic Regression → Spam/Ham prediction
TF-IDF converts the words in each message into numerical features, and those features are then used to train the Logistic Regression model to classify a message as either "ham" or "spam."
The interesting part came when I started looking at the model's decision threshold.
Initially, the threshold was 0.5. This meant the model would classify a message as spam only when its predicted spam probability was at least 50%.
I experimented with lowering the threshold to 0.4, 0.3, and 0.2.
And this showed me something important:
Lower threshold → more spam detected → higher recall → lower precision
At 0.3, the model caught significantly more spam while still maintaining good precision:
Accuracy: 97.39%
So 0.3 gave the best balance among the thresholds I tested.
The biggest thing I took away from this project isn't the accuracy, though.
It's understanding what actually happens between raw text → numerical representation → model → probability → decision → evaluation.

and the other interesting thing i noticed is that my first model achieved accuracy of 96.42% which feels very fantastic but which is not since my dataset was imbalanced. The dataset i used roughly have 87% ham and 13% spam. This mean A COMPLETELY USELESS model will simply predict "ham" every single time would still achieve around 87% without detecting a single spam message. Which actually helped me to see great accuracy might not always be great prediction

If you want to won a lottery, you have to make money to buy the ticket

-Nightcrawler

when you took it literally 😭

Episode 4 is finally here! 🎙️🔥

I sat down with Birhan Nega, a software developer with over 10 years of experience, entrepreneur, company founder and content creator.

We talked about his journey in tech, building a company in Ethiopia, working with international teams, discipline, remote opportunities, AI and what genuinely separates an experienced engineer from someone who only knows how to write code.

Watch the full episode here 👇

🔗 https://youtu.be/w8SHWwDehS0