Inspiration The inspiration for NexaGuard AI came directly from the "Trust Bottleneck" described in the Nexora Whitepaper. We realized that while the Workers Network allows founders to find talent globally, the actual workflow is still slow and manual. Founders waste hours reviewing code before releasing payments, and workers often wait days for their funds. We asked ourselves: "What if trust wasn't a manual process, but a line of code?" We wanted to build a system where quality work gets paid instantly, without human intervention.
What it does NexaGuard AI is an autonomous quality assurance agent that acts as a bridge between GitHub and the Nexora Blockchain.
Escrow Creation: A Founder deposits funds (in $NEXORA or USDC) into a smart contract linked to a specific task.
Automated Review: When a Worker submits a Pull Request (PR), NexaGuard AI wakes up. It reads the code changes and audits them using a Large Language Model (LLM) trained on security best practices.
Decision & Payment:
If the code passes: The AI Agent autonomously signs a transaction on the blockchain, instantly releasing the funds to the Worker's wallet.
If the code fails: The AI comments on the PR with specific feedback on what needs to be fixed, keeping the funds locked.
How we built it We built NexaGuard using a "Thin Frontend, Heavy Agent" architecture:
Blockchain: We wrote the NexaGuardEscrow.sol smart contract in Solidity and deployed it to the Nexora Testnet. It features a specialized agentRelease() modifier that only allows our verified AI wallet to trigger payouts.
AI & Backend: The brain of the system is a Python application using LangChain. We utilized Gemini 1.5 Pro for its large context window, allowing us to feed entire file diffs for analysis.
Integration: We used GitHub Webhooks to detect real-time code pushes.
Transactions: We used Ethers.js to handle the cryptographic signing of transactions on the server side, effectively giving the AI its own "bank account" and agency.
Challenges we ran into AI Hallucinations: Initially, the AI was too lenient, approving "hello world" code for complex tasks. We solved this by implementing a "Double-Check" prompting chain where one AI agent reviews the code, and a second AI agent reviews the review before signing.
Context Limits: Handling large Pull Requests was difficult. We optimized this by writing a script that only extracts the changed lines and relevant context rather than the whole repository.
Testnet Latency: We faced some RPC timeout issues on the testnet, which required us to build a robust "retry logic" into our transaction signer.
Accomplishments that we're proud of First Autonomous Payment: The moment we watched a GitHub commit automatically trigger a wallet balance update on the Nexora explorer was magical.
Security First: We successfully prevented the AI from releasing funds for code containing a simulated "Re-entrancy Attack," proving the auditor logic works.
Seamless UX: We built a dashboard that makes the complex blockchain logic invisible to the user—it just looks like a standard project management tool.
What we learned We learned that Agentic AI is the missing link for decentralized work. Smart Contracts are great for holding money, but they are "blind" to the real world. AI gives them "eyes." We also gained deep experience with Account Abstraction concepts, effectively treating an AI script as a first-class citizen on the blockchain.
What's next for NexaGuard AI Multi-Modal Support: Expanding beyond code to review design files (images/Figma) for the Creators track.
Dispute Resolution: Implementing a DAO-based "Human Appeal" system if a worker disagrees with the AI's rejection.
Mainnet Launch: optimizing gas costs for a full deployment on the Nexora Mainnet.
Log in or sign up for Devpost to join the conversation.