EthCC 2: Julien Bouteloup

hey guys can you hear me yeah cool okay so i'm going to give a talk a presentation of something we've been working on for the past two years it's called flying carpet we pivot many many many times and so the flying carpet is a decentralized geospatial layer for earth observation so the problem is like earth observation like data from space you got a lot of data from space available but the only way you can access data from space is by using massive corporate massive firm massive big institutions that give you access to data so you can buy the data but then it's very difficult to treat to analyze the data in order to extract data inside for building uh or for automating uh decision for example you want to calculate the production of coffee in brazil you want to do parametric insurance for farmers all this kind of stuff so what you can do you can actually extract that from space and then build data insights so the vision is to build a decentralized protocol that provides actionable insight from geospatial data using machine learning algorithm mechanical incent incentive mechanism and also smart contracts and then so those machine learning models can be decentralized and then anyone can actually buy or stake take a percentage of ownership of this model in order and then they can earn revenue so if if an investor or if someone use this model to extract data from space then they everyone that actually has a stake in this model will be able to make a revenue in income so we build it's a protocol for geospatial data and yeah the idea is to build financial instrument instruments on the top of this decentralized protocol so then you can actually build prediction market you can build proof of observation from space for example if you want to drive as i was mentioning insurance you want to if you want to make sure that something happened in a specific gps location so there is a catastrophe then you can trigger a smart contract that is attached to an oracle and the oracle is actually the the model that is responsible for analyzing the catastrophe is hashed into for example an ipa phase location and then you actually retrieve data inside from this specific model um so you can build different kind of financial instruments on the top of this of this uh protocol for example exchange market as i mentioned nozius or gear0x or private uh you can also build like dao so you can have like specific decentralized organizations that are responsible for climate change and i'm going to tell you uh present like a use case we've been working with microsoft and national geographic and you can have so adult different be able to buy a specific model to transform for specific use case um so this is different partnership that we've been working with so axa we've been helping a little bit on axon consulting for parametric insurance so they can actually use satellite data in order to extract useful information and then and then make actional actionable insights or decisions and nephilia is a reinsurance company you want the biggest free insurance company united states and actually worldwide so network application i was mentioning that the the idea is to interact the physical world so using smart contract and machine learning to make sure that you can actually use physical data so we've been working on five different use cases crops parametric insurance for in the insurance industry carbon offset for forestry industry rooftop anomalies so you can actually calculate and analyze rooftop from space and then you can you can calculate what is the percentage of anomalies and you can also plug this into prediction market we've been doing uh community trading for coffee in brazil and uh tracking climate change from greenland the melting speed of the of the ice so how we start the project we first started project using drones but then we had a problem of scaling drones we built the first one the first bounties network was using a tcr mechanism where you can actually incentivize people to go at a specific location and then gather data using their drone but the idea you need to use a specific sdk in a dji drone to prove that the data was collected at a specific location is actually accurate and 100 percent at this specific gps location so it's actually you can test it and um it's on the link alpha that's flying copper that network you can actually like pause the bounty and then get uh tell what kind of data you're looking to gather this is in london and uh this is on um on the wing company but you also have the magnet you can go and there's like uh we give you like some tokens to use the platform um so that's so yeah that's what basically what you can oh and then we built then the next phase we built a web 3 mechanical turk so the idea is to we gather a lot of data from space but it's very difficult to test the data you get from space so the idea is to incentivize people to go to the platform and then we send these images to multiple different people and then they have to stake they have to recognize okay this is a land of coffee this is not a line of coffee this is a land of coffee this is not the land of coffee and then they get paid so you basically incentivize people to to analyze to annotate your images so we did something with a company that is focused on on on forestry so the idea is to is to tackle the the market of carbon offset we're talking about the tree and market cap it's the idea is to basically prove that at this specific location you have a forest and this forest has been has been absorbing carbo um carbon so you build a model that is responsible for capturing data from space i mean we capture data from space for different satellite providers and then we feed those data into the model then this model is actually responsible for analyzing per square kilometer how much carbon offset was absorbed so if you generalize this and you scale this kind of use case imagine you could have trees in your garden and those trees you could be incentivized to keep those three trees alive because you'll be earning money from the european commission because those three actually participate in climate change and in order to do that you can trigger approval observation from space we collect data and then we feed the model and we give a binary output this is something we want we want a grant from for microsoft a national geographic it's a grant that specializes in ai for for earth so the idea is to we are sending people in greenland to so the we we build them we're building a model that can analyze the speed of the melting highs in greenland so if you have a model that can do that then you can actually plug this into a prediction market so you can say okay what is the speed of the melting ice is the speed of the melting ice in two years we're gonna be is going to be faster than the speed last year and you can actually build all this kind of financial instrument on top of it this is the one that we're currently doing with a big one the biggest hedge fund in coffee and this one is basically to analyze live production of coffee in brazil this is very cool i'm going to announce something at the end of the presentation what you can do with this kind of use case and this is the idea this design comes from 0x but the idea is to have multiple different models that are on the dashboard and anyone can actually stake inside and capture an ownership of this model and then if those model those models are being used then you actually earn money and then you can you can see how short-term medium or long-term uh investment if you want to if you want to bet on the market this is like using bonding curves i'm not going to go into that this is how you basically earn a percentage of ownership into the model but i think with sto is probably better and then we we just received a grant from nazis and the idea is to build uh on the top of nozis mercury version two is to build a prediction market for all those different models and we are going to start with the coffee coffee model so we are going to plug into nozis um a protocol and not protocol but like what if you're building in terms of smart contracts and then we bring in order for us to build the dashboard uh oh yeah sorry it was like wrong slide so yeah the idea is for example you can actually ask a question will the uh 2009 coffee yield in brazil exceed 2018 and then this proof of observation will actually be the oracle for this specific market financial market so this is how we you can actually find the link how we explain the entire proposal that we ship to nozis how are we planning to build this uh this one and now what we're trying to do is we're trying to um not be trendy but a little bit it's like basically get into the generalized mining because in order to run all those different models and to make sure that the data was captured and you can analyze this data you need many people to verify this data and you also need people to run this model in a decentralized way so we uh i can show you so let me see so we built something it's the first um the first step where you can you can basically like use the ui so you don't you you cannot run this validator on the cloud on the data center because you another data center maybe because you need the ui so we actually use google earth to double check that the information that was captured from satellites is correct and then we pay people so you can actually earn something like 120 20 die per hour which is like 120 u.s dollar per hour because we actually paying people at five dime per square kilometers and we been processing 3500 kilometers square kilometers of coffee in brazil and then we're going to extend this use case to many many different models and there is different people that we've been partnering with and then now we actually getting into european space agency because european space agency agency they really want to get into blockchain ai so we are going to partner with them like getting into consulting with them in order to build this kind of like communication across different satellites that can analyze data from space that's it you have questions oh yeah and the first validator that'll be running this generalized mining will be stack capital uh okay where is the data coming from is it coming from the european space agency and wouldn't be having decentralized data access to the satellite data be a prerequisite of all the other stuff you do give me give me a sec because yeah at the moment that's what we're getting the dollar from and also european space agency but the problem of this european space agency for example they actually do a specific partnership with airbus and airbus uh you need to have a specific partnership with them because the data you get from satellites it depends on you can get cheap data but it just depends on what kind of quality data you want do you want a sub meter 0.5 square uh resolution for the data or do you want something you don't care and you just want a three meters uh resolution so and then you also want do you want a dollar fresh or do you want a dollar you don't mind us about three months ago data so this is this is all these kind of different information you have to take into consideration and then we're using planet.com near mod initial cast and we also use isa but isa is not very fresh data at the moment for us we will need to partner with airbus for example and in terms of decentralization oh cool oh that's not just the one that's this one don't like this um yeah that's planets.com neomap usercast and in terms of decentralization this is really something we're looking into so now with the swarm of satellite you could potentially send your own satellite for cheap money and you can actually the idea is to basically decentralize the space i mean observation but if i say that it's a bit crazy and i've been talking to the guy behind you and he's got experience in a white block so he's got experience in satellite communications and also being building something they want to open source it's already open source i think and you can i mean potentially the idea is to build a flying carpet could actually help building a satellite as a service no it's not difficult we actually doing this but it's yeah we're doing this you got the platform online you can get the data but this thing is like how do you make if you for example if you're building a prediction market that would be responsible for uh that has 20 or 100 million at stake for making a proof of for making an oracle decision a binary a decision or even a scholar scholarly decision a specific prediction market you need to be 100 percent sure these that have never been tamper-proofed so you need decentralized way but the moment it's it's easier to it's probably better to use a satellite data than actually use drones data because drones i mean in order to hack a satellite it's i mean it's a lot of resources you know no oh yeah they've got they've got private keen and public key on the side tonight so you can actually do that i think he knows better than me but i was talking to um catapult and uh it's basically the space agency in the uk we were there like last week and they actually use um based on gps location they actually sign the data but the thing is like when you have different satellite that actually capture data it's very difficult to because that is not the same because it's not the same it's not when you're looking at you're looking at different angles from the pictures cool all right hey so i'm wondering how the ownership of the ama models work because it seems like if they're public that wouldn't be very good so they need to be like private in some way but how can they then be shared well this is something we've been talking with uh i mean isaac over there so yeah the thing is like the idea is to if you if you pay data scientists to build a model you don't really want the model to be open source but you want the model to be very be 100 sure that this model that extract data inside is always the same and never change so you can actually hash this model and then store this model in somewhere in a validator in a like kind of like an enclave so it's running but you cannot access the the the the code of the model but you can still give outside outputs i mean that are inside right since you're kind of relying on some service to keep the model available yeah i mean this thing is like you could say oh i want to use a golem and zig knox and then all this kind of stuff together to build to make sure the model is never being accessible from the inside and provide i don't really know the moment we're building model and the plan is to decentralize this model for anyone to yeah i mean we need to encrypt this model so we're talking to different people that can help us on that [Applause] you