12th June — Stopped Consuming and Started Building
Most days in tech, it’s easy to feel productive without actually creating anything.
You watch tutorials.
You save repositories.
You bookmark articles.
You tell yourself you’re learning.
Today was different.
Instead of consuming content, I decided to build something from scratch.
The Idea
I spent the day working on a Telecom Fraud Detection project.
The goal was simple:
Can we identify potentially fraudulent callers using behavioral and graph-based features instead of relying solely on traditional spam detection methods?
At first, it sounded like a straightforward machine learning problem.
The deeper I went, the more I realized how much thought goes into real-world fraud detection systems.
Beyond Simple Call Data
Most beginner datasets stop at basic features:
- Call duration
- Number of calls
- Caller history
I wanted to go a step further.
So I started exploring concepts used in modern telecom intelligence systems:
- Spam report counts
- Receiver block rates
- Sequential dialing behavior
- Reputation scoring
- Graph-based fraud associations
One of the most interesting concepts I discovered today was graph analytics.
Instead of looking at a phone number individually, you can analyze how it connects to thousands of other numbers.
A fraudster may change their number.
Their network behavior is much harder to hide.
Get Vanshika’s stories in your inbox
Join Medium for free to get updates from this writer.
That single idea completely changed how I thought about fraud detection.
Building the Dataset
Since telecom fraud datasets are difficult to obtain publicly, I designed a synthetic dataset containing:
- 10,000 records
- 15 engineered features
- Behavioral indicators
- Reputation-based metrics
- Graph intelligence features
The objective wasn’t just generating random numbers.
The challenge was making the data realistic enough for machine learning models to learn meaningful patterns.
Publishing on Kaggle
After creating the dataset, I published it on Kaggle.
What I thought would take five minutes turned into a lesson about documentation, usability scores, metadata, and community-driven datasets.
I learned that creating a dataset is only half the job.
Making it understandable for others is equally important.
A good dataset isn’t just data.
It’s documentation, examples, explanations, and usability.
A Small Realization
A few months ago, I mostly focused on learning syntax and solving coding problems.
Today, I found myself thinking about:
- Feature engineering
- Fraud analytics
- Graph networks
- Machine learning pipelines
- Dataset design
- Product presentation
That shift feels exciting.
Not because I know everything now.
But because I’m starting to think like a builder instead of just a learner.
Final Thoughts
Today’s work probably won’t change the world.But it’s one more step toward becoming the kind of engineer I want to be.
A little less tutorial watching.
A little more building.
And honestly?
That’s progress.
Building. Learning. Publishing. Repeating.
