
Verifiable AI on Blockchain: Spotlight on virtual and IRL winners of the Flare x Google Cloud Hackathon
Over the past several weeks, experienced developers from across the globe competed in two editions of the Flare x Google Cloud Verifiable AI Hackathon – an in-person event at UC Berkeley and a virtual one through DoraHacks. Participants were tasked with combining Flare’s enshrined data oracles with Google Cloud Confidential Computing to push the boundaries of blockchain and AI innovation. We’d like to take a moment to spotlight some of the standout projects.
Winners of the In-person Hackathon
2DeFi
🏆 Prize:
1st Prize: AI x DeFi (DeFAI)
2nd Prize: RAG Knowledge
📑 Project Details:
2DeFi, the user-friendly AI-powered platform that simplifies the Flare onboarding process, took home two prizes at the Flare x Google Verifiable AI Hackathon! Built by Alex Lu and Hitharth Khurana, 2DeFi is a Gemini-powered platform that helps TradFi users transition to DeFi on Flare through Google login, portfolio analysis, and guided investment strategies in TEEs without requiring crypto expertise.
The key benefits of 2DeFi are:
- Easy onboarding: Google Sign in for a ready-to-use, embedded Flare wallet that is created immediately upon signing up.
- Guided investment strategies: Users will receive a custom DeFi strategy after signing up. The strategy features staking, liquidity pools, swaps, and memes.
- Integrated DeFi: 2DeFi utilizes major platforms on Flare, such as Sceptre for staking sFLR & Blazeswap for Liquidity pools and swaps.
- AI-powered portfolio analysis: By leveraging Gemini models, 2DeFi can scan your Robinhood portfolio with 1 picture to analyze your portfolio and assess risks & assets.
- Automated support agents: Users can query the agent for real-time help with executing DeFi strategies and to gain insights on the Flare ecosystem.
Flare Fact Checker
🏆 Prize:
1st Prize: Social AI Agents
📑 Project Details:
Flare Fact Checker tackles the immense challenge of pervasive misinformation online, which can often prove incredibly difficult due to the sheer amount of information available. Created by Luka Skeledzija, Mark Znidar, and Gregor Kržmanc, Flare Fact Checker is an independent AI and RAG-based verification tool for scientific claims on Farcaster that provides correctness scores, explanations, and sources through a network of TEEs.
The key benefits of Flare Fact Checker are:
- Verifiability: Verifier nodes run in a TEE, which prevents the modification or tampering of the knowledge base and language models.
- Correctness score: Each verifier node returns a “correctness score,” a short explanation and list of sources that either supports or refutes a claim.
- Natural language UI: UI components are generated via natural language commands.
- Decentralized access: Any interested team or individual can set up a verifier node with its knowledge base. Every claim that is input for verification is then verified by several nodes in an attempt to reach consensus. Verifier nodes are rewarded using Shapley values of each node.
Command Flare
🏆 Prize:
1st Prize: RAG Knowledge
3rd Prize: Consensus Learning
📑 Project Details:
Command Flare captured our attention with its AI-driven interface that allows users to execute transactions, swaps, lending, and borrowing with simple, conversational commands. Built by a UC Berkeley team – Ravi Riley, Rohan Patra, and Vardhan Shorewala – Command Flare is a stateful, AI-driven interface for Flare that combines Graph RAG and Vector RAG in TEEs to provide accurate responses, enable transactions, and generate UI components through natural language commands.
The key benefits of Command Flare are:
- Conversational simplicity: Users can execute transactions like swaps or lending just by giving natural-language commands, such as “swap my tokens.”
- Flare insights: Gain access to insights on Flare through semantic search.
- Contextually accurate AI responses: The Letta module retains conversational context for personalized and continuous user experiences.
NFT Deep Appraisal
🏆 Prize:
1st Prize: Consensus Learning
📑 Project Details:
NFT Deep Appraisal, created by first-year UC Berkeley Electrical Engineering and Computer Sciences (EECS) students Roger Mas and Mohul Shukla, is a consensus learning system that evaluates NFT values by combining appraisals from AI models within a TEE using confidence-based weights to improve accuracy over single-model approaches.
The key benefits of NFT Deep Appraisal are:
- Confidence-based weights: NFT Deep Appraisal assigns higher influence to models with consistent and coherent reasoning, which is guided by a 70/30 price-consistency ratio.
- Multiple AI models: Combines outputs from Gemini, Claude, and Llama to produce a blended appraisal.
- Secure environment: The project runs in Google Confidential Space TEEs with vTPM attestation.
- Real-time data: NFT Deep Appraisal queries the Moralis API for up-to-date info on the AI output analysis
Winners of the Virtual Hackathon
The virtual edition of the Flare x Google hackathon also brought together teams from around the world who yielded equally impressive results. The winners of the virtual competition demonstrated how off-chain verifiable AI can reshape user experiences, from DeFi to social media. The first project we’d like to highlight is Quince Finance.
Quince Finance
🏆 Prize:
1st Prize: AI x DeFi (DeFAI)
📑 Project Details:
Quince Finance introduced voice-controlled DeFi to make Flare interactions seamless and beginner-friendly. It leverages Google Cloud’s Confidential Computing and Flare’s oracles to maintain usability and security. By harnessing voice commands, Quince lets users stake FLR or provide liquidity without friction.
The key benefits of Quince Finance are:
- Easy onboarding: Users can sign in with their Google account to access a wallet that is securely stored in a TEE.
- Voice-controlled DeFAI agents: Give spoken commands to perform DeFi tasks such as staking, swapping, or managing liquidity. You can give simple commands like “Swap 50 FLR for SGB,” completely hands free.
- Secure & confidential: Quince uses vTPM tokens for authentication, which adds a layer of protection for user data on the platform. It also incorporates TLS and dual Google ID verification to keep user data and funds safe.
- Simple dashboard: Quince offers an easy-to-use dashboard and interface that make Flare DeFi more accessible to a higher number of users. See all of your balances on one page.
ScribeChain
🏆 Prize:
1st Prize: Social AI Agents
📑 Project Details:
ScribeChain is an X bot built by Liam Pereira that summarizes X Spaces and provides answers to follow-up community questions operating from within a TEE.
Each summary is automatically generated and in concise, bullet-point format. All that’s required is for a user to simply type “ScribeChain summarize” under a post with an X Space link, and ScribeChain will download the audio, then use Gemini 2.0’s multimodal capabilities to post a bullet-point summary as a reply.
The key benefits of ScribeChain are:
- Summarizes audio: Users no longer need to re-listen to whole Spaces to be caught up on any information shared or discussions held. Simply type “@ScribeChain summarize” under a post with an X Space link for the highlights.
- Follow-up questions: ScribeChain also provides answers to any Spaces follow-up questions from the Flare community. If a user has a question about something in the summary, they can respond to that summary with a question.
- Flare Integration: ScribeChain aims to connect with Flare’s TEEverifier contract to deliver on-chain proof of AI operations to boost trust and transparency.
See the full list of In-person and Virtual Hackathon winners!
Each of our hackathon winners displayed fresh, innovative solutions that combined AI, TEEs, and DeFi to make blockchain more intuitive and accessible. The projects highlighted here are just a taste of all the things that can be built when leveraging Flare’s unique infrastructure.
For a full list of every in-person and virtual hackathon winner, please visit the DoraHacks page: https://dorahacks.io/org/flare