Introduction

Okay, thank you very much, Alessia, for the introduction. And hello, everybody. I'm Pablo Romero. I'm the CEO and co -founder of Graniot Satellite Technologies.

And I'm here today to show you how we transform satellite imagery into an operational AI workflow.

What Graniot Does: Earth Observation for Real-World Change

So, what do we do at Graniot? We do exactly what you are doing right now. You are observing the Earth.

So that's what we do with remote sensing techniques, with remote sensing technologies. We observe the earth and try to identify what is changing.

This is our mission, this is our purpose.

Defining Earth Observation Beyond Space

In fact, earth observation is defined by the European Space Agency as the process of acquiring observation of the earth, surface and atmosphere via remote sensing instruments.

So we cannot say that earth observation technologies is only space. Earth observation technologies

can be for example UAV, drones, helicopters, vehicles that have a camera that observe the earth.

A Brief History of Earth Observation

In fact the earth observation is not something new. Earth Earth observation was born already in 1858, thanks to the curiosity of a French photographer

that wanted to see what is happening in his city. It was not Paris, it was another city. And he decided to, went into, in a balloon, and take the camera with him to take photos

of the surface of the city.

Thanks to him, we can talk today that the first satellite was launched in the space by the United States of America in 1960, and this satellite was the TIROS -1. You can see also the satellite of 1957 from the Union Soviet, but it was not an Earth observation satellite, it was just communication satellite.

From Imagery to Business-Ready Intelligence

So what we try to say from Granite is that Earth observation technology is not only about looking at Earth, observing the Earth, it is about understanding what is changing on it, what is changing on the surface related to agriculture, related to

vegetation, to critical infrastructure, to your own city, and in fact this is what what we do at Graniot. We transform this earth observation technology data into

business -ready intelligence for companies that manage overall two specifically sectors, vegetation and infrastructures. Here in the left you can see different type of services that we were building in the last four years.

Key Use Cases: Vegetation and Infrastructure

Some of them are, for example, crowd monitoring. It's like the same that that Giovanni Di Mambro with Elysian are doing, but instead from the earth, from the space.

So we get data from these crops, from images taken by satellites.

We do also analysis about the risk of deforestation from agricultural areas outside of the European Union, because products such as coffee, cacao, or soy, cannot be imported into Europe, into our countries, if they proceed from non -deforested, if they proceed from deforested areas. So we do measure also that these areas are non -deforested.

And also we can control how the construction of a road is progressing quickly, quickly, thanks to images of very high resolution, because we can go from 30 per 30 meters per pixel to 30 per 30 centimeters per pixel. It was a military technology a few years ago, now it's a civil technology, and we can use it also for doing visual inspections, for example.

And lastly, but very important, we do also, for example, tree inventory. There are some customers that ask for us, how many trees do I have in this parcel, in these farms, in this city? This is able today, not only do it by drone, but also by satellite imagery.

The Pain Point: Manual Validation at Scale

But in this specific case, we had a problem, and the problem is here. This is me at 2 a .m. in the morning, validating and checking the results of some AI models, machine learning models, about the tree counting. counting.

So we needed to count the trees manually, let's say, after getting the first results of our inputs, of our scripts. But the problem was not only this one. It's that

in the team, we are a team of seven people, I was not the only one doing this. In fact, I'm the CEO. It was the CTO and another developer. So it's three resources, three human resources checking and validating something for our customer before delivering the final result, before delivering the final report.

TileGen: Turning an Internal Tool into a Product

So that's why we decided to create and use AI, also lovable, for creating an internal tool. This internal tool was TileGen. And it was an internal tool to help us to count the trees in an automatic way. way, to check in and validate the results directly using generative AI. After checking this solution, checking this validation and so on, we said, okay, if it's helping us, why not could help our customers?

So today, thanks to this event, we are launching the first beta testing program for TileGen, and now it's becoming a product also for our customers and this product it is

Core Ideas: Divide and Conquer, Knowledge Is Power

based in two concepts divide et impera so divide and conquer and sciencia potencia est sorry for the for the pronunciation but power knowledge is And or knowledge is power,

power and why because in imaging when you have a satellite imagery or drone on imagery, it is very important what you can send, which is the image that you send to a specific algorithm.

So it is not the same sending this quantity of images to the same model than sending all these images, smaller images. It depends on the model you want to use, tree counting, field boundaries, it depends on that.

Why It Works: Generative AI + Remote-Sensing Models

Because TileGen is born with a mix of generative AI that you may know, NanoBanana from Gemini, GPT -MH2, MetaSan Model 3, Flux .2, and pre -trained and trained models for satellite imagery and UAV that we created at Granite in our four years and a half of life.

So it's a mix between generative AI and specific models we created for our customers. That's why it's so powerful and valuable.

Shifting from Consultancy to SaaS

So with this solution, it is also a transformation of what we do at Graniot. It's like going from a consultancy company, where I count the trees, I create the report or whatever, to

to a real SaaS business model because it is the user getting inside the platform and doing everything by themselves.

In fact, this is a strategy. This is what we all do every day with ChatGPT, with Cloud, and so on. But in this case, professional profiles.

We are able to save more than the 60 % of the time of these professional profiles. These professional profiles are GIS, geographical information system users, technicians, that are using, checking images every day.

And it is cheaper because you don't have to pay for a consultancy product. If you want, you can do it, but first do it yourself and check the results.

Product Walkthrough: TileGen Demo

Let's see the demo.

Scenes, Data Sources, and Upload Options

Okay, here you have TileGen. It is a very easy and and an easy user interface for the professional profile, where you can see different scenes.

The scenes is the image of our customer. So now we move also from, if you don't have the image, I can provide the image for you from the satellites, but if you have your own drone image, do it by yourself.

You don't have to send us that image, just do it by yourself. You can upload the image to the to the to the platform and then you have active

jobs and these active jobs are the different processing models we were building in the last years. So for example, here we have a scene from a customer.

Okay, got it. This customer uploaded its own image. In this case it is not

not even an image coming from a drone or from a satellite. It's a Google Maps screenshot.

Tiling and Model Configuration

So with Google Maps screenshot, here we have the tiles. That's why TileGen. The tiles are the number of images you are dividing the image.

And then you have already an auto -suggested tile configuration for this specific processing mode. This processing mode is field parcels. So we want to get every specific agricultural boundary from that specific image.

So, okay, I like it. I like the configuration. I can also adjust the image and the brightness, the gamma, contrast, and all of this. This is for professionals. And I can select what I want.

Okay, I want to process all these images, all these tiles. I can process a new job, or I can add it to an existing job. So if I create a new job, okay, I don't want to read you anymore.

Processing, Results, and Validation

The job has been created, and here we have every, okay, don't want it. Here we have every small little image from that specific image that we had before,

and it's processing automatically this is the mix between the generative AI and our models for that specific model fields field parcels so here you have the first results and we are just analyzing smaller pieces of that big area that you sent at the beginning and here you can say you can see that

Of course, we got every specific agricultural field from the image to the vector. So we can already work as professionals with that specific geometry in a map. And we can validate the results by overlaying the real image with the results that we got.

If we don't like it, okay, do it by yourself. Retry the task. Let the AI redo the task.

Map Outputs, Downloads, and Reporting

again so once all the little images are processed we can polygon eyes or we can see everything on the map okay on the initial map and if everything is correct you can download all these images and you can also create reports about okay how many fields did you get from all these images how which is the dimension

of these fields can you do any correlation between blah blah blah blah by yourself so just for showing you I'm out of time for sure here we have for

example the scenes and the subset by inferior okay I mean it is in testing It's in trial, that's why maybe we have some things.

And here we have the model of field parcels. So we can show in the main map, in the main image, the results of our model, as you can see here.

So it's like, let the user do everything, like we do with ChatGPT, but be careful because you are consuming credits, okay? OK, and since you are professional, you know better what you are doing instead of us.

Conclusion and Beta Program Invitation

And that's why we decided to create a product from an internal tool. And I think this is one of the best validations that we can have as a company.

And just for ending my presentation here, so that was everything. But since I said at the beginning,

Meaning we are launching the first beta program. We are looking for people who try, test and validate our tool and give us feedback.

So if it's your case or you know GIS professionals, companies, whatever, please send the result of scanning the QR code that you see on the screen and it will help us a lot during this month. Thank you very much.