GemFund

🚀 Inspiration

Trust is breaking down. We’ve seen the headlines evolve from fabricated stories—like the infamous Kate McClure homeless veteran scam that stole $400,000—to the thousands of fake donation links appearing minutes after the Turkey-Syria earthquake using recycled footage from old disasters.

The statistics are alarming: GoFundMe reports over $10 million in fraud attempts annually, and nearly 1 in 10 online campaigns shows signs of deception.

This creates a "Donor's Dilemma": People want to help, but they are afraid of being scammed. Furthermore, traditional donation platforms are slow, heavily taxed by fees, and opaque—donors never really know if their money reached the destination.

As a team of developers passionate about social impact, we asked ourselves:

"What if we could build the world's first crypto-native donation app that uses AI to automate due diligence in seconds?"

GemFund was born to answer that question. We believe transparency shouldn't be optional—it should be programmed into the money itself.

💡 What it does

GemFund is the first mobile crowdfunding platform powered entirely by Cryptocurrency and secured by Gemini 3.

We solve the two biggest problems in charity today: Fraud (via AI) and Opacity (via Blockchain). Unlike traditional platforms, GemFund assigns a real-time "AI Trust Score" (0-100) to every campaign.

Think of it like a "Credit Score" for charity. We use a Defense-in-Depth approach:

• Multimodal Fraud Detection (Powered by Gemini 3):
  • 👁️ Visual Forensics: Gemini Vision scans campaign imagery to detect stock photos, deepfakes, and recycled footage (e.g., spotting that a "current" flood photo is actually from 2018).
  • 🧠 Semantic Reasoning: It analyzes the campaign story for high-pressure tactics, manipulative language, and logical inconsistencies common in scams.
  • 🔄 Cross-Modal Verification: It compares the Visual Evidence against the Text Claims to ensure they match (e.g., ensuring the weather/location in the photo matches the written description).
• 🔗 Blockchain Forensics: Because we use crypto, we can check the creator’s wallet history to detect "Wash Trading" (fake donations to boost visibility) and suspicious new account activity.
• 📜 Smart Contract Escrow: Funds are not held by a company. They are locked in Ethereum Smart Contracts, ensuring transparency and enabling automatic refunds if fraud is detected.

🛠️ How we built it

We built GemFund as a robust ecosystem connecting Web3 with the latest generation of AI:

• The Mobile Experience (Flutter): We chose Flutter to build a high-performance, cross-platform mobile app. This allows us to access native device features (like the Camera for live verification) to enhance security, providing a smooth experience for donors on the go.
• The Brain (Gemini 3 Integration): This is the core of our innovation. We leveraged Gemini 3’s advanced reasoning capabilities to process disparate data types simultaneously.
  • We engineered a complex prompt chain that feeds Gemini a multimodal context window: Raw Image Bytes + Campaign Text + JSON Blockchain Data.
  • We use Gemini 3 not just as a chatbot, but as a Reasoning Engine to perform logical deduction (e.g., "Does the visual data support the text claims? Is the wallet activity suspicious?").
• The Ledger (Ethereum & Solidity): We wrote custom Smart Contracts to handle donation logic. By using Cryptocurrency, we eliminate cross-border fees and ensure an immutable, public record of every cent donated—something fiat currency cannot do.
• The Backbone (Supabase): Used for managing user sessions, Edge Functions for OSINT tasks, and secure storage.

⚡ Challenges we ran into

• Bridging Mobile and Web3: Connecting a Flutter mobile app to Smart Contracts while simultaneously calling heavy AI endpoints was a latency nightmare. We optimized this by using Supabase Edge Functions to handle the AI processing asynchronously in the background, keeping the mobile UI snappy and responsive for the user.
• AI False Positives: Initially, the model would flag legitimate campaigns as fraud just because the grammar wasn't perfect. We solved this by Fine-Tuning our System Prompts to focus on factual inconsistencies and visual evidence rather than writing style.

🏅 Accomplishments that we're proud of

• ✅ Successfully building the first fully functional crypto-donation mobile app integrated with Multimodal AI in a hackathon timeframe.
• ✅ Leveraging Gemini 3's multimodal nature to solve a real-world financial problem, going beyond simple text generation.
• ✅ Integrating 5 layers of forensics (Text, Image, EXIF, Blockchain, OSINT) into a single, easy-to-understand "Trust Score" interface.
• ✅ Creating a system where the AI acts as an autonomous auditor, reducing the need for human moderation by 80%.

📚 What we learned

• Context is King: AI is powerful, but it needs the full picture. We learned that feeding Gemini just the text or just the image wasn't enough. The magic happens when Gemini analyzes Image, Text, and Crypto-Wallet data together to find contradictions.
• Trust requires UX: Blockchain and AI can be intimidating. We learned the importance of hiding the complex tech behind a simple, friendly mobile interface so that non-technical users can donate via crypto with confidence.

🔮 What's next for GemFund

• "Proof of Life" Challenge: Implementing a feature where Gemini challenges creators to upload specific real-time photos via the app's camera (e.g., "Take a selfie holding a spoon next to a clock") to verify liveness and prevent bots.
• Cross-Chain Support: Expanding to Layer 2 chains (like Polygon or Arbitrum) to drastically reduce gas fees for small donors.
• NGO API: Releasing our "Trust Score Engine" as a public API so other charity platforms can use our fraud detection technology.

Built With

• dart
• gemini
• github
• supabase
• typescript