I started to type these lines on the train back home, after a full immersion at Pycon Italia 2024. I have been working with Python for 7 years now and up until some days ago I have never been in person to a Pycon. Could you believe it?

Honestly, for at least two years, I had the excuse of COVID19 stopping in-person events. Actually I have even presented at EuroPython directly from my living room. For sure, that was a cool experience but it missed the great thing about conferences: meeting and talking to people or, how smarter people would say, networking.

For personal reasons I was at the conference only on Thursday and Friday. But even skipping the last day the conference was excellent.

And on top of all of that, as casually mentioned in the title, I have also been a speaker.

Talks

Over 100 speakers using Python for ML, web apps, pure backend, anything you could think of, in the same place. Choosing what to follow was one of the hardest tasks of the conference.

I have attended to talks close to my day to day job, in particular:

• How you can write a dataframe-agnostic library by Marco Gorelli’s, core contributor of both Pandas and Polars, on his library Narwhals that is able to deal under the hood with both Pandas and Polars dataframes.
• How to Handle Very Large Data with Python by Jill Cates, senior data scientist at Shopify, comparing Pandas, Polars and DuckDB. Pandas does often its job, but when you have to deal with larger datasets you should move to either Polars or DuckDB.
• Scaling Machine Learning Training and Inference by Ilyas Chaoua, senior data scientist at Vedrai, showing how they manage the training and inference of their machine learning models.
• Computer Vision IRL: dall’idea al deploy on-premise by Egon Ferri, computer vision engineer at Immobiliare.it, explaining how they deploy their image recognition models using PyTorch and FastAPI.
• AI on a Microbudget - Methods of Machine Learning Miniaturization by Katharina Rasch and Christian Staudt, data scientists and freelancers, outlining how it’s possible to reduce the size and inference time of models by using transfer learning, quantization, and distillation.

Other talks that I enjoyed are:

• Speeding up Whisper for speech recognition by 10x - Optimisations and walkthrough by Vaibhav Srivastav, ML Developer Advocate at Hugging Face, explaining how using different techniques they were able to build a model derived by OpenAI Whisper that has comparable accuracy and it is way faster.
• functime: a next generation ML forecasting library powered by Polars by Luca Baggi, ML Engineer @xtream and functime core developer, demonstrating how this library can be fast and effective.
• RAGtime with Haystack! How to build a RAG with Haystack 2.0? by Tommaso Radicioni, Data Scientist at Ai Know You, explaining how Retrieval Augmented Generated works.

Anyway since the talks were recorded I plan to catch up the ones I missed and learn even more stuff. They were so interesting overall that even on Saturday when I was not in Florence, I watched a couple of them in streaming.

Events

You could say that you could get the same info following the conference online, but in this way you miss the events. Take for instance Thursday with PyBeers after the evening keynote (completely agree on the fact that developers can have a real impact on the world based on the company they choose to work for). I really liked the short talks about abnormalib and on Bayesian marketing by Emanuele Fabbiani. And then PyDrinks (you could see how risky it could be to have a talk the next day) in the context of a really cool atmosphere. Accompanied with food and drinks, it was nice to talk about Python libraries, projects and personal interests. At some point I ended up at a table trying to learn the card game of Cabo.

I am a little bit jealous of who got to attend to PyDinner! But what makes great events are of course people organizing and participating.

People

A bunch of people from Italy, and of them a good number working abroad, from all around Europe and even overseas (not a little fact since it was overlapping with PyCon US).

I met a university classmate that I haven’t seen since graduation, a principal engineer coming from my very city and now living in the Netherlands, data people working in my same industry (that let me tell you is quite a niche).

I ended up talking with numerous people indeed of Python, how it is used at work and in personal side projects, cloud architectures and machine learning. But also many non tech stuff, a subsample in random order: how it is for an Italian to live in another country, lasers for microscopes, how it’s to be a data scientist freelancer, Japanese language, Chinese different alphabets and keyboards, best Canadian food, a comparison of wheather among Milan-Zurich-London, how to pronounce correctly th sound in Brazilian.

Being a speaker

Can you imagine presenting one of your personal projects at such a conference? Not only the thrill of having an audience and talking about something you built, but also some nice perks, like free conference tickets and a pre-event for speakers-only. And it makes easier also to start a conversation since people may want to know more about your project/talk.

Thanks to a good dose of luck, my talk “Exploring Art with Python: Building an Italian Art Bot” was selected. During the presentation I showed my experience in building a Python bot that posts images of Italian paintings on the social platform Bluesky. I will publish soon a post focused on the project, for now here it’s a preview:

Considering that in my same time slot I wished to attend to a couple of other presentations, I would say having 40 people listening to my project was fantastic. Moreover friends and family could enjoy the show on streaming. At the end of it, I received many questions, several words of appraisal and ideas for adding features, and the bot’s followers increased by 10% (using percentage instead of absolute numbers, you are doing it great. The followers before the talk were 29).

One thing in common with my EuroPython experience was issues with the microphone, but that’s just my reduced practicality.

What I liked less

I just described how fantastic this conference was, but let’s add a little about what I did not like so much:

• Having missed in person Vicki’s Boykis keynote on Saturday afternoon! I followed the keynote from home, but it was hard to get an autograph from there.
• So many great talks going on in parallel that you could not attend all of them.
• Some of the rooms were so full that no everyone wishing to have a seat could attend (you could follow on streaming though).
• The conference was so great and packed with talks and events that you had basically no time to visit Florence (but it was cool to have the option to join afternoon guided tours).

You see the level of the conference was so high that the negative points are basically small side effects of the quality of speakers and organizers.

Take-aways

What are my take-aways from this conference, besides the awesome t-shirt and the stickers?

• So much information and inspiration for my daily job just talking with people about their work and challenges.
• A bunch of libraries that I discovered during the presentation and I want to try out.
• Booking the accommodation well in advance: it would have been cheaper with the discount for the conference staying at the four star hotel hosting it than in a small studio.
• I should attend to more events related to Python and data in my region. Actually I should add it to my list of objectives for the next 12 months.
• The community is freaking great!

I am looking forward to the next PyCon!