We Won ETHGlobal NYC. Here’s What Actually Made the Difference.
This was my second ETHGlobal. Last year I went to the venue to check-in, didn’t talk to anyone, went home to build by myself, ran into a lot of issues, tried to get help through discord, almost got scammed (only response I got were from bots telling me what I’m testing doesn’t work because I don’t have enough $ in my wallet — luckily I was so noob with crypto, I couldn’t figure out how to fund my wallet, and by the time I got a friend to transfer me some $ hours later, the bot/scammer’s account got deactivated), couldn’t get anything demo-able to work, didn’t realize I needed to present to be considered for any prizes… needless to say, failed in all possible ways. However, I got to chat with some sponsors at the end of the hackathon, they actually liked what I was trying to build, wished I’d came to ask more questions in person to provide my feedback and get my issues resolved, looked into the scammer bot with me and apologized about the bots have more legit looking profiles than actual admins in the channel. I enjoyed connecting with the community and walked away feeling hopeful that the next time, I will do better.
This year, Andre Chuabio and I built GoHealthMe over the weekend, and won Best Usage of Chainlink Confidential AI Attester and $2k in prizes (our biggest winning yet!) Here are our learnings.
The Project
The idea: health version of goFundMe — get paid to be healthy — get paid in USDC the second once it’s confirmed.
Anyone can chip into your pool. A parent, a friend, an employer, insurance companies, the government, even a stranger can stake USDC behind your goal (or against). Hit the goal and the money’s yours; miss it, you lose your stake, and the $ rolls back to the pool / split between people taking the other side of the bet. Maybe this is the motivation you need to finally lose that extra weight, drop your cholesterol, bench 200 lbs, or just get 8 hours of sleep.
The fun part is verifying a health claim without exposing someone’s personal data (easy wearable intergation or upload a file) using Chainlink’s Confidential AI Attester (basically a secure enclave), spits out a signed verdict, and that verdict decides the payout on-chain. The raw health data never touches the chain. World ID keeps it one real human per goal, and Circle/Arc handles the USDC.
What Made the Difference This Year
1. Staying at the venue and talking to All the sponsors
I was determined to talking to EVERYONE and really take advantage of all the sponsors this year. We went in not really knowing what we are going to build, so we went around asking the sponsors what their product does, what their bounties are and what they want to see built (especially the ones with free merch, I was determined to get the $350 I spent on claude code back in free food/merch — also tip do NOT use claude code api, just get max)
Once we started talking to all the sponsors, we wanted to use ALL of them, everyone was so passionate about their product, they didn’t mind at all that we had no idea how their product work and didn’t have a clue what we wanted to build. We were seriously considering incorporating all the sponsor tech and presenting to all of them (but we read the rules a bit more closely and realized we are technically only allowed to present/submit to 3 of them — but maybe next year).
Sponsors we Submitted to:
- Chainlink: (our fav) We spoke at length with Darby Martinez from Chainlink — who showed us his project EASEeHealth and we took a lot of inspirations from it.
- Blink: We got a lot of encouragement from the team who are also fellow Canadians and some of them went to the same college as me (we decided to build for their prize pool because they were such great hype men).
- Unlink: we decided to go for the joined private track by unlink / dynamic / arc to maximize amount of sponsors we can submit to — but unlink wasn’t really there most of the time
- Dynamic: ran into some issues integrating with MetaMask and apparently one of them couldn’t sleep thinking about the bug we were seeing — appreciated their help
- Arc: we really liked their agent market place concept and wanted to launch something on it, but when we told them our idea they didn’t think it’s a good fit — so just used as a chain — maybe another time
Other Sponsors:
- World: was cool to set up our worldID, adding it had some challenges but they were there to help us debug live — we saw a similar app on their app, and thought that it’d be interesting to try to get published there
- Uniswap: loved talking to their team and loved the concept of lego pieces for developers to build with, but our idea is too far from de-fi / swapping to use any of their api/hooks — would like to explore building some hooks for fun ways to distribute fees etc
- 1inch/Aqua: loved the Aqua team, they told us the story of building this product in this month long city together, too de-fi for us
- Google Cloud: lots of onchain data using GCP — but we didn’t need pricing data
- Canton: loved how passionate the representative was, if we had more time would’ve used it — liked how it has privacy baked in (similar to what unlink offered butinstitutional facing)
- Privy: super easy to integrate with, but we wanted to qualify for the joint prize by unlink/dynamic/arc, which ended up being a bit of a pain to integrate with — should’ve just used this instead
- Li.Fi: sounded really cool but we didn’t have a use case for it
- Hedera: we loved the idea of launching it on different chains — would’ve added this if we had more time
- ENS: we tried to get this to work but had issues creating ENS names
2. Chainlink: CRE + Confidential AI Attestor + AI Skill
Neither Andre or I knew much about web3/crypto going in, so our strategy was to build something at the intersection of web3 (just enough to qualify for prizes) and AI (which we know well).
- Chainlink’s CRE made it really easy for us since it obstracted away the web3 part and made everything typescript, which is much more familiar for us — somehow it made all the smart contract concepts seem easier to understand too.
- We also really liked the Confidential AI Attestor concept since I work in health tech and Andre and I had been building a bunch of health tech centric hackathon ideas where data privacy and confidentiality is at the core of all of them (hippa, pii, etc).
- Chainlink’s Agent Skill: this was the cherry-on-top, this made developing with claude so much easier
3. Building what we know + what we think they’d want
We initially spent time making sure our core idea worked — and once we got that to somewhat work, tried to add in some pieces to expand the AI attestor:
Model Performance benchmarking: we mocked different files and scored the different AI models on their accuracy
Get Nikki Hu’s stories in your inbox
Join Medium for free to get updates from this writer.
Model Simulations Evals with Veris — expand to run thousands of simulations on these scores and further fine tune the models
Additional Features we’d like to explore — Multi-modal support for video evidence — how do we verify if someone can bench x?
4. Working with Andre — team work makes the dream work
The is honestly the most important one. Having a good team really makes magnitudes difference! He’s gotten so efficient at creating new projects (he has different agents and skills set up for different pieces of the workflow, from ideation and planning, to spinning up a frontend, deployment, adding database/auth, etc)— I learn so much from him every time!
Andre is also a huge sports fan. He had plans to watch the Brazil game + Knicks game with friends, but he missed both because we were still debugging. Towards the end of the night, he had the Knicks game on a third screen and everyone at the hackathon crowded around us cheering for Knicks as they made championship history. He also skipped celebrating with friends after so he could wake up at 6 am to finish our demo for the 9:30 am. I’m so glad his sacrifices paid off — always inspired by his dedication and focus.
(Andre is currently open to work — if you’re looking for a great AI engineer, definitely hire him.)
Whats Next
We’ve got a backlog of projects we want to actually launch — follow along and DM/Comment to try these out, or let us know what you want to see built:
- GoHealthMe — hit a verified health goal, get paid instantly in USDC. Anyone can stake into your pool; miss the goal and it rolls back.
- MediGuard (won) — HIPAA-compliant data leak protection middleware that redacts patient PHI before it touches any LLM, plus a voice onboarding agent that triages patients straight to the right specialist. No forms, no GP appointment just to get a referral. Live on PyPI.
- rehabAsCode — clinician-in-the-loop rehab protocols. Patients chat with an AI coach; any protocol changes go through a clinician dashboard for approval before going live.
- Merit — turns your research and open-source work into O-1A /EB1-B visa evidence. Tracks all eight USCIS criteria and exports an attorney-ready dossier.
- Healthmaxx (personalized video wellness coaching via Tavus avatar + Apple Watch data)
Full list on Andre’s site.
Join Arya in our hackathon in July
The best way to learn and keep up with the latest in AI is by building things. There’s a great community of builders in New York and we’re really excited to connect with and learn from more of them.
I’ll be hosting a health-tech focused hackathon in July, more deets to come, come build and meet the team at Arya — especially if you are active in the job market, or if you just love meeting and connecting with other builders, we are actively hiring engineers who are excited about building in the AI era! Check out careers page: https://aryahealth.ai/careers
