Data from the Physical World

⚠️ Important Note:
PyData Amsterdam is transitioning to Luma. ✨

• Please subscribe to PyData Amsterdam on Luma: https://luma.com/pydataamsterdam
• And register at the Luma for this event: [https://luma.com/6fuljd88](LINK)

***
What’s up, PyData Amsterdam community! For our next meetup, we are bringing you

From greenhouses to construction sites, some of the most interesting data problems in physical AI show up in the real world. That's where data has to turn into decisions and movement.
​This meetup brings together Monumental, Source.ag and PyData Amsterdam for an evening of educational talks from engineers who work on these problems every day. Each talk is about 30 minutes, with plenty of room for questions.
​Schedule

• ​18:00-18:30 – Walk In + Welcome
• ​18:30-19:00 – “Data problems in the greenhouse” with Sebastiaan Vermeulen (Data Scientist, Source.ag): This talk digs into why forecasting a greenhouse harvest is such a stubborn ML problem. A single growing season gives you maybe 40 data points per crop, so there's none of the "just add more data" luxury most ML talks assume. Half of what data does exist is collected by people walking the rows with a tablet, which brings all the usual mess of human-recorded data. On top of that, plants don't play by the rules a standard model expects: a normal ML model will happily predict fruit before flowers, or ignore that fruit takes six to eight weeks to actually grow.
  You'll see how Source deals with that by blending hand-built biological models with learned components, so the model knows what a plant can and can't do before it ever sees the data. There's also the less glamorous half of the problem: getting a forecast accurate and consistent enough that growers actually build their week around it. Expect real code, real failure stories, and a good look at what it takes to make ML trustworthy outside of a spreadsheet.
• ​19:00-19:30 – “Robot Whispering on the Construction Site" with Josefine Quack (Forward Deployed Robotics Engineer, Monumental): Imagine a lot of carefully designed, millimeter-accurate systems operating in one of the most hostile environments imaginable: a construction site. Dust and mortar everywhere, temperatures swinging, rain, snow.

Josefine was there when Monumental's robot fleet consisted of just a handful of systems; it's since grown to more than 100 robots operating simultaneously across different sites, in different countries.

In this talk, she'll give a quick introduction to how the systems operate, how the team continuously deploys new features onto a live, running fleet, and how you debug a system that's more than 100km away from you.

• ​19:30-21:00 – Dinner, drinks and good conversations. We'll provide the drinks (non-alcoholic too) and dinner - and rely on you for good conversations! See you there!

​Details

• ​Address: Plantage Middenlaan 62 1018 DH Amsterdam
• ​When: 8 October, 18:00-21:00

​Who should attend?
​Data scientists, software engineers, Python users, and anyone curious about how data works in the physical world.
​About the Speakers
​Josefine Quack
​Josefine is an experienced Forward Deployed Robotics Engineer at Monumental, where she hacks and plays with robots all day. As a result those robots have already built over 100 houses. Please don't leave any chocolate around her unattended.
​Sebastiaan Vermeulen
Sebastiaan Vermeulen leads harvest forecasting at [Source.ag](https://source.ag/?utm_source=luma) (econometrics/quant finance background, 4.5 years in). He'll unpack why predicting greenhouse harvests is a hard ML problem: sparse, messy real world data, biology pure ML gets wrong, and the mechanistic/ML hybrids Source uses to fix it.
​About the hosts
​Monumental builds autonomous robots that construct buildings. We're already laying bricks on family homes, canal walls and industrial buildings across the Netherlands!
​More about Atrium, our operating system for software-defined construction
​Open roles @ Monumental
​**[Source.ag](https://source.ag/?utm_source=luma)** builds AI for greenhouse growers. We turn the flood of data coming out of modern greenhouses into clear, daily decisions, so growers can produce more food with fewer resources.
​More about [Source.ag](https://source.ag/?utm_source=luma)