Skip to content
Blog

From New Media to data science

I studied New Media at CityU's School of Creative Media. Here's how that led to data science and AI products, and why a creative media background still shapes how I build.

3 min read

I studied New Media at City University of Hong Kong's School of Creative Media, in a double degree with Digital Media at Leuphana University in Germany. I didn't plan to end up in data science. But looking back, the path makes more sense than it felt like at the time.

Creative media, then systems

New Media sits where code, design, and interaction meet: generative art, game development, HCI, interface design, animation, and a fair amount of computer science. Somewhere between writing p5.js sketches and building a small survival game, I realized the part I enjoyed most was the system behind the visuals: the rules, the data, the logic that makes something respond.

My graduation project, Deepstory, made that obvious. It was meant to be an artwork, and it was, but most of the work was data pipelines, model training, and debugging memory leaks. I enjoyed every part of it.

Data, then AI

A summer at the Hong Kong Census and Statistics Department, building a lakehouse pipeline, was my first real data job. Internships at BASF moved me further into analytics: NLP candidate scoring, recommendation systems, forecasting models, Power BI. Almost five years at ALTEN in Hamburg built the fundamentals: ETL pipelines, semantic models, governance, my first production RAG systems, and a lot of stakeholder conversations. Then DHL Data & AI: Document AI, multi-agent systems, and GenAI tools in production. Throughout, I've co-built HKLawyer.ai on the side.

Alongside work I did an M.Sc. in Management & Data Science at Leuphana, which filled in the theory I had been picking up on the job: statistics, Bayesian modelling, optimization, and a thesis on keeping LLM agents grounded over long tasks.

Why the creative media background matters

Insights should be easy to get at, not buried in spreadsheets. Creative media taught me to obsess over clarity: what someone sees first, what they ignore, what they remember. That's the same question a good dashboard or a good AI answer has to solve.

It shows up in small decisions every day:

  • A Power BI report is a piece of interface design. If the important number isn't the first thing you see, the report has failed, however correct the DAX is.
  • A RAG answer is a piece of writing. Citing sources and explaining things in plain language matter as much as retrieval precision.
  • A model nobody can use is a prototype. Deepstory only became real once it had an interface someone else could operate.

If you're making a similar switch

A few things that worked for me:

  1. Build things end to end. A small project that goes from raw data to something someone can click teaches more than ten notebooks. It also gives you something to show.
  2. Learn the boring fundamentals. SQL, data modelling, and pipelines are where most real projects live. The fancy models sit on top.
  3. Use certifications as structure, not as proof. Studying for exams like PL-300 or DP-600 gave me a map of what I didn't know. The projects are what convinced people.
  4. Write as you go. My Deepstory dev log started as homework. It turned into the best record of how I learn, and this blog continues it.
  5. Keep your old lens. Whatever you studied before is an advantage, not a gap. For me, it's caring about the person on the other side of the screen.