Inspiration Insurance today is slow, bureaucratic, and adversarial. The traditional "file and wait" model leaves users anxious and often unpaid for weeks. We were inspired to build Vetrox Agent—an impartial, autonomous entity that replaces paperwork with code. We wanted to answer the question: "What if an insurance company was just an AI that paid you the second it saw the damage?"
What it does Vetrox is a decentralized, AI-powered claims processing agent that operates 24/7.
Instant Analysis: Users upload evidence (e.g., a photo of a broken phone). The Vetrox AI Agent scans the image instantly to identify the item and assess damage severity.
Automated Payouts: If the damage is clear, the agent directly interacts with a Smart Contract to trigger an instant crypto payout to the user's wallet. No humans involved.
DAO Governance: If the AI is unsure or rejects a valid claim, the case is escalated to the "DAO Governance Console." Here, human administrators can review the evidence and override the AI's decision to force a payout.
Trust Scoring: The system tracks user trust scores and policy statuses on-chain.
How we built it We built Vetrox using a hybrid Web2/Web3 architecture:
The AI Brain: We utilized Google Gemini (via the API) for its multimodal capabilities. It analyzes image evidence and outputs structured JSON data regarding the damage type and severity.
The Backend: A Python (FastAPI) server acts as the bridge. It handles the image uploads, prompts the Gemini model, and manages the logic for approval.
The Blockchain: We wrote Solidity smart contracts deployed on the Sepolia Testnet. The backend uses Web3.py to sign and send transactions automatically when a claim is approved.
The Frontend: Built with React and Tailwind CSS. We implemented a "cyberpunk" aesthetic with a custom "Eye" animation that tracks the user's cursor to give the agent a sense of presence. We used Reown (AppKit) for seamless wallet connection across multiple chains (Base, Linea, Ethereum).
Challenges we ran into
The Oracle Problem: Getting the AI's off-chain decision (Web2) to trigger an on-chain transaction (Web3) securely was difficult. We had to create a secure backend wallet acting as an oracle.
Prompt Engineering: Tuning Gemini to strictly reject ambiguous images while approving clear damage took several iterations to prevent fraud.
API Security: As seen in one of our dev screenshots, managing API keys securely in a hackathon environment was a good reminder of the importance of environment variables!
Accomplishments that we're proud of
Achieving a fully completely automated flow: Image Upload -> AI Analysis -> Blockchain Tx in under 10 seconds.
Building the DAO Dashboard, which provides a safety net for the AI, making the system practical for real-world use.
The UI/UX design—specifically the animated "Watcher" eye that reacts to the system state (scanning vs. idle).
What we learned We learned that Generative AI is surprisingly good at acting as an insurance adjuster for physical goods. We also gained a deeper understanding of transaction management and gas estimation when automating payments on EVM chains.
What's next for Vetrox Agent
Video Analysis: Upgrading the agent to analyze video feeds for more complex claims (e.g., car accidents).
Multi-Agent System: Having separate AI agents for fraud detection and damage assessment to vote on a claim.
Mainnet Launch: Deploying the smart contracts to a mainnet L2 like Base or Linea for lower gas fees.
Log in or sign up for Devpost to join the conversation.