Yes, we are here to tell you not only about Chainlink’s confidential AI attestor, which Andre will be demoing soon , but also how to use Chainlink in your projects at this hackathon. So, this is the most important QR code for you—it's the presentation QR code . Everything is there : bounties and all the tools we use. Yes. And I'm Solik, and you— and yes, my name is Andre.
I think you already know me. I will be here already know me. I will be here at the booth to help you with help you with any technical questions or if you need help with product ideas or anything else. So don't be afraid to approach the stand. Yes.
These are also my contacts. This is Yes. So, first let's talk about Chainlink bounties. First Chainlink bounties. First , we really want to see how you use CCIP, and the three best projects using these workflows will receive $2,000 each.
Another bounty is connecting the world with all other Chainlink services. If you want to use other Chainlink services, the top two projects will receive $1,000 each, and this is a new feature we have today. Andre, are you going to do a demonstration soon? The best use of Chainlink's confidential AI attestator—the top two projects will receive $2,000 each. And also, you know that we are giving you the opportunity to continue your project.
This is also from Chainlink Global. If you have a project and want to add Chainlink to it, the top two projects will receive $1,000 each. And what else is important to you? Let's talk about the skills of a Chainlink developer. So if you're a developer, install the Chainlink developer skills, agent skills, we have skills for all of our Chainlink services.
So, you're ready for this, and like I said, the Chainlink runtime we're talking about. These are all the links you need to use CCIP. We won't talk about that right now because we're really excited about the Chainlink Confidential AI Attestor demo, so Andre, it's up to you. Thank you. Thank you.
Speaking of Chainlink's confidential AI attestor, as you saw, we have a special bounty for it, and we will have a special guest judge for this track. If you are familiar with Chainlink and the Chainlink ecosystem, you may know who Professor Rules is. He will join us on Sunday to help us with the evaluation of projects in this track. So, I just wanted to emphasize that . So before we start the demo, what is Chainlink Confidential AI Tester?
I don't know how many of you have tried to get a loan, buy a car or an apartment, but you know that if you want to get a loan or verify your finances, you can't do it privately at the moment. I mean, you essentially need to disclose documents, involve notaries and lawyers, right? These notaries verify the document and confirm that your words are true. Imagine that you want to confirm your financial status to get, for example, a car loan. But in DeFi we can't do that.
We thought about this problem, and our research and engineering teams created this tool, or, let's say, an experiment. This is an early alpha version, there isn't even documentation yet , but we decided to present it at a hackathon and see what you can create. So, all the use cases, like unsecured DeFi unsecured DeFi loans—that's exactly the case I'm going to demonstrate today. Anything related to KYC, AML, proof of reserves, anything that involves uploading documents and analyzing them—and I'll show you that in a minute—is a great candidate for this direction, where Andreas will join us again as a judge. I'll just show this slide; this is a preview, it's like a playground URL.
And after that, I'll go straight to the demonstration. You will see not only a playground there, but also API documentation, which is not as complete as what you are used to on the Chainlink website. So you need to follow the link. Also, an unsecured lending demo project on Foundry with this Chainlink Confidential AI Tester that I'll show you Tester that I'll show you , and AI skills. Because we decided that you don't need to become an expert at working with Chainlink Confidential AI Tester over a weekend; you can just use skills.
We have skills for everything. So, the main rule for skills: focus on use cases. Skills have experience building with chains, so just use them in any environment: cloud, code, cursor, codex or any other, they will figure it out themselves. So, without further ado, let's follow this link. I hope you see that.
Maybe I should zoom in a little . So what I'm saying is that if you follow this link, there's this playground there. To work with Chainlink's confidential AI with Chainlink's confidential AI tester, you will need an API key. So if you want, you need this API you need this API key. Just come to our stand.
We will send you API keys. Write to us on Discord in the Chainlink Partner Channel. We will send you API keys. Just a note, you need an API key. Let's say, for the Let's say, for the under- collateralization demo, we want to prove that we have a strong financial base.
This is a fictitious PDF generated by AI for a specific person. So I want to download this PDF and ask a Chainlink confidential tester if this person is suitable for a DeFi loan. The thing is, I don't want this document to become public. Here's what the Chainlink Confidential AI Tester will do: as you can see, it will run this analysis in a TEE, which stands for " trusted execution environment," and more specifically, the AWS Nitro Enclave. I have a request here.
So I'll just copy and paste it. Essentially, based on these financial documents, is this person a suitable candidate for a loan, and the rest is just the format of how I want to get the answer. The thing is, when I get certified, I can verify it on the blockchain. So I'm going to save this decision in a smart decision in a smart contract. If I go to the code, it's already deployed.
As you can see, the workflow is quite simple. I will upload the documents for the AI to analyze. Once inside this confidential AI confidential AI tester TEE, instead of me getting the solution directly, I wait for a callback from Siri. Siri will actually decrypt the solution and store the attestation directly in this smart this smart contract, which is extremely simple. The idea is to expose these few viewing functions that my hypothetical DeFi hypothetical DeFi protocol can use to confirm that I can get a loan.
OK. So, without further ado, we have a request based on the financial documents provided. Is this person a suitable candidate for a 500 USDC unsecured loan? I am going to provide these documents. You can download them from download them from any URL.
You can even use the curl command. So you don't need to, you'll see it in the repository. You don't have to use this interface. I just want to highlight it because it's beautiful. You can just use curl or any other program.
Um, I'm uploading a file, which uploading a file, which is actually this PDF that you just saw, and I need a Siri callback instead of me receiving it directly. So I'll go back to my demo, and in that repository I'll run the command. First of all, I'll check my version of Siri. If you have Siri version 19 or higher, this feature is available to you. If not, please update the version, especially those working on the continuity track.
The function previously worked with simulations where you had to provide, for example, an HTTP trigger. You needed to provide it as a payload for the trigger, but now, with this version, I can compile the workflow and you see the message here: waiting for HTTP waiting for HTTP request to run. So, on localhost 2000/trigger it waits for a log it waits for a log trigger or an HTTP trigger. This makes it much easier to run simulations based on your Khan feedback, as many of you have run your own shell scripts and scripts to do so. Now you can just open it.
And since I can't just make a request to localhost 2000, I'll use Ngrok to just give it a dummy URL that I can pass here for the callback. So, this is localhost 2000. We need the trigger slash. So, trigger trigger slash. So, trigger is an endpoint.
Just make sure Just make sure you add it. So you add it. So , this is a trigger. And I can just send a request. As you can see, this is the JSON of the request.
So if you run the curl command at that address, it will do the same thing. She immediately replied " immediately replied " accepted". And we see accepted". And we see that it is in the queue. I can check the status.
Processing is underway. I can check the status again. Um, and while the processing is finishing, we should see a real simulation here with on-chain action on the testnet. Because if you haven't noticed, after the simulate command I added a broadcast flag, which will actually run the on-chain will actually run the on-chain command on the test network, even though it's a simulation. So if you want to work with an on-chain runtime, you don't need to deploy the workflow to production, although we can help you with that, but a simulation with the broadcast flag is enough to allow you to do end-to-end testing.
Here we see a cache of the transcript of this attestation, recorded along with a digest of the document. We see that the approved parameter is true, meaning the loan is approved, and here we see that the entry into the blockchain has occurred, as there is a transaction hash. If I go back to this smart contract and find the internal transactions, we should see a new one. Um, yes see a new one. Um, yes , 21 seconds ago.
And if I look it up, it should be the same BB10 transaction hash, which means if I query this contract for the last solution, and the payload was just number one, let's say, in a configuration where the borrower address borrower address is some kind of fictitious one, we should see the last solution, which we do. So, we see that this account shows significant monthly income, meaning it all comes from the results of the LLM analysis. The LLM used for this demonstration was Llama 4. You can also switch to Qwen 3.6.
But all of this was run inside an AWS Nitro Enclave. That is, it all works in a trusted execution environment execution environment . You can use other options, of course other options, of course , but it would be ideal to provide a document or documents for analysis. If you have any questions, we can discuss them now. Yes.
Do you want a microphone? Part of Siri. Yes. So, let me clarify that Siri is not a mandatory requirement for use with Chain and Confidential AI Tester. Chain Confidential AI Tester can work without Siri.
It was just convenient for me to post this certification on the blockchain via Siri. Siri worked, yes, actually, everything was local. What you see here is a local simulation, as I ran a Siri workflow simulation here with the --no-broadcast flag. I want to point out that the entire repository is available if you follow this URL that I am sharing with you here. This unsecured loan example is the exact repository I was running.
Here you will see all the steps and settings. The first scenario is a local simulation with older versions of the Siri CLI, where you essentially need to provide a payload, and I have a dummy example in the repository. But I did scenario 2, which was a complete end-to-end process. Instead of running this after opening port 2000 on localhost: instead of running this curl command, I ran this through the interface to show you. But it works perfectly with curl, and it works with any programming language that can do this " that can do this " under the hood": fetch for TypeScript, I don't know, Go, REST, Rust, whatever.
So it's perfectly fine to use this without Siri. Okay, are there any other questions you might have ? No. Yes. Oh, good question.
So, the question was, how to verify that the URL is not fake after the callback? Let's see where the payload is payload is . Here you have, or maybe even in get, you have all the...if, I don't have the ID, you have all the details about all the station resources, when they were created, etc. We don't have an actual endpoint here, so I understand why that was a really good question, but anyway, this is an alpha anyway, this is an alpha version for this hackathon, so you can't test it through the code, if you know what I mean, because that data isn't available here, but I understand what you mean.
Yes. Good. Are there any final questions about the demonstration? Everything is fine. OK.
Class. That was good. Then we can gradually complete this workshop. So, once again, this is the QR code for the presentation. If you joined later, you will find all the links there.
Again , these are our awards. So, four awards for this hackathon. You can combine them. One is for expanding your previous Chainlink projects. We have already received several questions from you.
So yes , that's perfectly acceptable. We are familiar with what you are building. So continue in the same spirit. New—these are the top three projects using the Chainlink runtime. That is, Siri, including all the privacy features we showed you earlier.
They are still relevant. We don't have a separate track anymore, but it's still part of CRE. So it's perfectly acceptable, you can do it. The two best projects using any using any other Chainlink service. Whatever you find in the documentation or elsewhere—just do it, it works.
And the two best projects using what you just saw, Chainlink Confidential AI Tester, reminding us that Jewels will be joining us as a judge for this track only. For Chainlink Confidential AI Tester, you will need an API key. Find us at the stand or write to us on Discord or through partner channels. I will post all the links and a short message with everything you need a little later. If you want to work with Data Streams, one team has already reached out to us; we have some ready-made test credentials.
So again , come by the booth or message us on Discord. We will provide you with test credentials for Data Streams. So, you don't need to imitate them anymore. So you can just work with them. And, well, actually, I wish you success.
And if you have any final questions, Saul and I will be here to answer them now or after the workshop. Thank you after the workshop. Thank you . Thank you.
Automatic transcript — names and jargon may be misspelled.
