Messy vCards with Pure Python, Pandas and Polars

The Problem That Started It All

It hasn’t occurred to me in the past to write an article on a problem I was just solving. Today was different. In what may be my last code contribution with District 4 Labs, I felt encouraged to write about this particular experience. It began with a database dump that looked like a digital archaeological site. Buried inside a field called value was an assortment of contact details stored in vCard format — names, titles, prefixes, emoji-filled nicknames, and the occasional vCard tag salad.

The mission? Clean this mess. Extract just the useful data name, phone, email, company, position, etc from a sea of structured chaos.

But cleaning up isn’t the only concern in the data world. Efficiency matters too. So, I decided not only to build a solution but to benchmark it using three different approaches: pure Python, Pandas, and Polars.

Phase 1: The Straightforward Python Attempt
The first approach was a clean and readable pure Python script. Using csv.reader and some well-placed string manipulation and regex, it looped through each line, parsed the vCard string, and extracted what looked like what I was looking for, name, phone, email, etc.

Pros

• Easy to write and reason about.
• Zero dependencies.

Cons

• Slower as dataset size increased.
• Lacks built-in optimizations for columnar operations.

Time to clean a 20MB dataset: ~2.45 seconds

Phase 2: When in Doubt, Call Pandas
Next up was my all-time trusted data wrangling friend, Pandas. I brought in DataFrame, used groupby and apply, and did some gymnastics with lambda functions.

It worked — but it wheezed.

Pros

• Powerful for structured data.
• Excellent ecosystem and documentation.

Cons

• Memory-heavy.
• Slower due to single-threaded execution and object-based operations.

Time to clean a 20MB dataset: ~12.02 seconds

Phase 3: Enter Polars, the Silent Speed Demon
Polars is the Rust-powered DataFrame library that feels like Pandas went to the gym and started eating clean. I didn’t know Polars until recently from a Data Engineer colleague. I was impressed the moment I read the introduction on their website.

With lazy evaluation, native multi-threading, and optimized memory usage, the same operation that made Pandas sweat ran like lightning in Polars.

Pros

• Super fast.
• Built for modern hardware and large data.

Cons

• Still maturing ecosystem.
• Smaller community than Pandas (for now).

Time to clean a 20MB dataset: ~1.31 seconds

Results Recap: The Showdown

Lessons Learned

1. Don’t underestimate pure Python. It may not win races, but it shows up.
2. Pandas is great, but not always performant. Especially on larger, string-heavy datasets.
3. Polars is the future. If you’re working with large-scale data pipelines or cleaning up gnarly datasets, it’s a game-changer.

Closing Thoughts
Data cleaning isn’t glamorous, but it’s where real-world projects live and breathe. Choosing the right tool can make the difference between an afternoon well spent and one spent watching your laptop fan spin like a Merlin engine.

If you’re wrangling vCards or any structured text data at scale, give Polars a try. Your CPU will thank you.

— -

Got questions or want a copy of the scripts? Drop a comment or connect with me on GitHub. Let’s clean up the mess — fast.