DEV Community

Cover image for From Drive Tests to Data Pipelines: Why I Went from RF Engineer to Data & AI
Pathum Dilshan
Pathum Dilshan

Posted on Originally published at pmpdilshan.github.io

From Drive Tests to Data Pipelines: Why I Went from RF Engineer to Data & AI

Originally published on pmpdilshan.github.io.

Two years ago my working day looked like this: drive-test logs, KPI dashboards, alarm lists, and a lot of Excel. Today I'm an RF and network planning engineer at Mobitel, reading for an MSc in Data Science & AI at the University of Moratuwa, and I build Python and Apache Airflow pipelines that process over 10 million KPI records a day.

I didn't plan a "career switch." I just kept running into the same wall, and data science was the way through it.

The wall: too much data, too few hours

A mobile network produces a staggering amount of data. Every cell reports dozens of counters every hour. When I started, first as a Huawei technical executive deployed to Hutch and then at Mobitel, most of my time went into collecting and formatting that data, not understanding it.

So I started automating. A Python script to merge exports. Then a scheduled job. Then a proper pipeline. Every hour I saved went into the part of the job I actually enjoyed: finding out why a cell was misbehaving.

What carried over from RF engineering

  • Thinking in systems. A dropped call is never one thing. It's coverage, interference, handover parameters, load and hardware all interacting. That's exactly how you need to think about a machine-learning model and its data.
  • Distrust of averages. Network engineers learn early that a "good" network-wide KPI can hide terrible individual cells. Data scientists call this aggregation bias. Same lesson, different name.
  • Domain knowledge. This is the big one. Knowing that a KPI jumped because two vendors define a counter differently, not because something broke, is worth more than any algorithm.

What didn't carry over

  • Statistics, properly. I could read a KPI chart, but I had to relearn probability, hypothesis testing and validation from the ground up.
  • Software habits. Version control, testing, reproducible environments. Engineers write scripts; data teams need systems.
  • Being comfortable with "it depends." RF has clear specifications. Machine learning is full of trade-offs where the honest answer is "let's test it."

What I'd tell another engineer

  1. Start with your own problem. Automate one report you hate. You'll learn more than from ten generic tutorials.
  2. Your domain is your advantage. There are many people who can train a model. Far fewer can tell you whether its output makes sense in a live network.
  3. Build in public. Write about what you learn. My Medium articles on medical AI and reinforcement learning taught me more than the projects themselves, because explaining forces clarity.
  4. Formal study helps, but it's not the starting line. I started the MSc after I'd already been automating for a while. The theory landed far better because I had real problems to attach it to.

Where this is going

My MSc research is on drift-aware anomaly detection for mobile network KPIs, a problem I met first as an engineer and now get to study properly. It's the clearest proof I have that the two careers aren't separate. The best work happens where they overlap.

If you're an engineer thinking about data and AI: you're not starting from zero. You're starting with the hardest part already done.


Pathum Dilshan is an RF & Network Planning Engineer at Mobitel, Sri Lanka, and an MSc Data Science & AI student at the University of Moratuwa. He writes about telecom, data and AI. Connect on LinkedIn or visit pmpdilshan.github.io.

Top comments (0)