Empowering Sexual Assault Survivors Through Cryptography | Samantha Ouyang - Zuvivor, FranklinDAO

Okay, hello everyone. My name is Samantha Ouyang. I'm a 21- year-old nomad, currently on a gap year from UPenn M&T, and I've worked in crypto since I was 16. And well, since last October, when I spent two months in Thailand in several temporary cities, I've been working on something called Survivor, a platform that empowers survivors of sexual abuse to report their abusers by connecting victims of the same perpetrator, so they know they are not alone and can find a way forward together. So, a little bit about my personal history.

I've been traveling to crypto conferences since I was 17, and the first time I was sexually harassed was when I was in Austin for South by Southwest that year. That basically made me hate the city ever since . But I ended up returning to Austin two years later for Consensus. This was around May of last year, and I started working on this idea in honor of my younger self. And at Megazoo, in Edge City in Thailand, I was inspired again to develop this project because I had to report my Airbnb host for sexual assault in Cambodia just before arriving in Chiang Mai.

And I myself was very afraid to report it. And the figures also reflect this. 75% of sexual assault cases go unreported . Victims are afraid to speak out for fear of retaliation afraid to speak out for fear of retaliation , defamation lawsuits, loss of reputation, and all these other reasons listed here that create what is called the first whistleblower disadvantage. So Survivor works to eliminate that disadvantage by allowing survivors to create a timestamped record of their assault and be connected with other victims of the same perpetrator.

Victims can submit a report where they complete as many identifiers as they know about the perpetrator, including name, email, social media, features, location, etc., document what happened, upload evidence, and verify their identity or KYC using ZK email or World ID. This is essentially to prevent false accusations or defamation. I also successfully implemented GPT-4 when it first came out, in order to describe an uploaded image and thus process the image data sent in the reports. Secondly, based on Cursive technology based on Cursive technology , the system leverages multiparty computation, or MPC, specifically the intersection of private sets, to identify overlapping information in submitted reports without revealing anything further about their identity.

And from here, a network of victim nodes is developed from the data superimposed on the connections. And so, this forms a knowledge graph structure . And well, groups will eventually begin to form around specific identified perpetrators as more pieces of the puzzle emerge from the stories of different victims to form a more complete picture. And survivor nodes can gain access to the network of other survivors in the group. And finally, once victims are part of a group victims are part of a group , they can check their perpetrators to see how many reports have been filed against them and connect with other victims of the same aggressor.

The platform also notifies victims when these matches occur and offers them the option to connect in an anonymous private chat to find support and solidarity with each other. This is just a small demonstration of the implementation of GPT-4o, which is basically able to identify elements in an image and then translate them into text. Here's a small demonstration of the front-end. Essentially, you fill out a form to report a perpetrator; All fields are optional, but the idea is to fill in as much as you know about that person. And in the end, you choose a verification method.

Here I am using World ID to essentially verify that I am a human submitting this report that I am a human submitting this report . After that, you can return to the original page, click on submit report, and that will be processed on the blockchain. I also used Arbitrum Stylus to write my smart contract, which was used to store data about report structures and related functions, as well as helping to identify and count overlapping reports. After Devcon, at Zuzalu, I started working more on the graph architecture I mentioned earlier. This was a small whiteboard session I did a while ago while I was working out the details.

First I worked on training a graph neural network using PyTorch for the matching algorithm. Different fields such as eye color, license plate, or phone number are encoded or integrated to become features of the nodes. Then features of the nodes. Then , the graph neural network learns these representations in order to identify or classify whether these nodes represent the same perpetrator or not. And now, with this network and using Ezekiel And now, with this network and using Ezekiel ...

how curious, I ran into the CEO of Ezekiel this morning when I went to Torchy's Tacos. It allows you to demonstrate the reliability of the model and that you are acting in good faith without having to reveal any of the confidential data sent by the victims. So now I'm going to take a quick tour with some sample data. This is the information that is essentially collected through the form and, as you can see here, although these two reports are not exactly the same with the same strings, it is noticeable that they probably refer to the same perpetrator, given that it is a John Doe and a John D, and both, one is Hispanic and the other is Mexican, and it is happening in a similar area. And then, that generates a higher confidence rate and you can probably say it's the same person, so it puts them in the same group.

And here's a similar one too, except this one's for Jane Smith. Um except this one's for Jane Smith. Um , you can see that she is white, um, one is Caucasian, and both are located in NYC or New York. Um, and lastly, here's a slightly atypical case. So, report four, um, has nothing in common with any of the previous ones.

And then, you end up with a chart structure that looks something like this, um, where those that are similar are grouped together, um, and those that, uh, are different are grouped into a different one. Um, and well, here I am, uh, running the program. And, essentially, it creates vectors of different, um, feature encodings of nodes. And at the bottom, it generates, uh, borders if they are similar enough. Um, and well, these are bidirectional edges well, these are bidirectional edges .

And, based on the graph above . And, based on the graph above , uh, you would see that zero and one are linked, and then one and zero are also linked. And then two, three, and three, two are linked. Um, and if you align the array on the edge indices to one side, you can see that, uh, that's the case. And they also see that the number four doesn't appear anywhere because it has no borders linked to it because it has no borders linked to it .

Um, here's a bit of the tech stack I used to build this. Um, and yes, here's some of the traction I've gained so far. So, since I started this project in October, um, it's gained a lot of traction which I'm very grateful for. Um, it has received great support from people who are very eager to bring it to their own communities. Um, I've received interest from people in communities like modeling, jiu-jitsu, and even the club scene in Berlin, and many more.

Um, and I started working on this around the time the news about DWF Labs came out. Um, and from there I shared my personal story about what happened in Cambodia, and also why I was working on this project. Um, and then that tweet kind of went viral as the crypto community really came together in support of raising awareness about sexual abuse and safety at networking events. Um, and my story was seen by many notable figures whom I have followed for a long time, such as Laura Shin, the famous podcast host, and also the executive producer of Vitalik's film. Um, and then at ETHCity, I ended up winning four awards at the hackathon as well, including first place on the Arbitrum track, and also the community's choice award, where they basically evaluated each finalist project on stage, and the one that received the loudest cheers won, and it was really moving to have a whole crowd cheering at the top of their lungs for me.

Um, and after that, the Arbitrum judge interviewed me, um, and that got posted on Twitter, retweeted and quoted a lot, um, I felt a lot of support from there, um, and after that, people started sending me money to my EVM address, which was really crazy. Um, and in the bottom right, I think you can see, I told my dad about this, that someone gave me money, and he said, oh yeah, basically joking about how I shouldn't let my investors down. Um, and yes, after Thailand, I'm following up with many of the ecosystems that were really interested in supporting, and looking for different grants. And I'm also looking for someone, like a technical co-founder or anyone who wants to help with this project. Um, and yes, thank you very much for your time.

It's been an absolute pleasure speaking at ETH Denver this year, um, and let's unite to hold the perpetrators accountable. Thank you.