How small do you feel, looking at this image? Makes me feel like my brain can't even process what it's seeing. What you're looking at is one of the first public images from the Vera C. Rubin Observatory, released today, the 23rd of June, 2025.

Among the releases were some beautiful shots of nebulae, showing the messy yet awesome process of star formation. But something else grabbed my attention, and that was the image of the Virgo Cluster.

It's a galaxy cluster in the constellation of Virgo, and this image puts its immense scale into perspective. These are thousands upon thousands of galaxies, all packed into a single, 3,200 megapixel frame that my iPhone 13 wouldn't even know how to look at.

What makes this observatory so special is its field of view. It's designed to scan the entire southern sky every few nights, generating an insane 20 terabytes of data each night. To put this into perspective, that's roughly the expected download size of Black Ops 7, except this data is probably more useful.

You don't have to look close to see what's going on in the top right. Yes, that is indeed three galaxies playing intergalactic tug of war.

You can see the awesome tidal tails streaming between them. The galaxy on the left is being pulled into the middle one. The one in the middle isn't so lucky. It's being pulled from both sides. And the one on the right, which looks like the largest of the three, is doing most of the pulling. We are watching a much larger galaxy forming, one we won't be alive to see finish, but it's still damn cool to see it happening.

Their reddish hue tells us that these galaxies are populated by older, cooler stars. They're in the final stages of a merger that's probably been going on for hundreds of millions of years. The bright, concentrated core of that main galaxy is what's particularly interesting.

A little trivia: astronomers believe that nearly every large galaxy hosts a supermassive black hole (yes, including ours). Now typically, they're dormant. They don't do anything, don't get up and find a job or pay rent. They just do... nothing. But sometimes, every so often, they finally get up to... eat. They begin feeding on the debris, gas, and stars within their host galaxy. When they do, they're considered "active," and we say the galaxy has an Active Galactic Nucleus (AGN). When they're active, the accretion disc (that awesome, swirling stream of hot plasma you saw in Interstellar) emits so much radiation that it can outshine the combined light of its entire host galaxy.

Why is this exciting? Those three galaxies in the top right are prime candidates for reigniting dormant supermassive black holes. But the keyword is candidates. Not every interaction funnels enough material right into the black hole's path. Regardless, these sorts of mergers are perfect targets to begin doing spectroscopy on. They scream "AGN".

But what about the others, where their characteristics aren't so obvious to the human eye? Or, more importantly, what do we do when we have such a massive amount of data from Vera C. that we can't possibly classify every galaxy? Astronomers can't inspect every single one of the billions of objects it'll catalogue. Spectroscopy, while incredibly accurate, is slow.

This is where it gets really exciting for tech.

In 2022, a paper from Ziting Guo and his colleagues at Yale showed that a Convolutional Neural Network (CNN) can be trained to identify likely AGN host galaxies based on their appearance (or morphology) with a high degree of accuracy. So, when the Rubin data is made public, we can feed these new images into similar neural networks.

In fact, I built one myself! It was a great learning process, going back and forth with AI to understand the tools, syntax, and the physics, but the result is a working model. I've documented everything, and you can check it out here.

The model has a recall rate of 91% for finding AGN. Its overall accuracy is 83%, but for this task, correctly identifying as many true AGN as possible is more important. A 91% recall means that if we feed it a batch of galaxies, it will correctly flag 91% of the actual AGNs in that batch. This lets us confirm them much faster.

These neural networks can act as a powerful filter, sifting through millions of images to flag the most promising candidates. They can learn to spot the signs of an AGN that even we might miss. The reddish colours, the central bulges, and so on.

Of course, this in no way replaces the work of astronomers, not even close. This is triage. It simply tells astronomers, "HEY. I'm, like, 91% sure that this is an AGN. Might be some exciting discoveries here!"

This is a new chapter in exploring our universe. These images are just a taste of what's to come. A whole lot of data is coming soon. I can't wait to see what we'll do with it.