AI-powered, on-chain hackathon evaluation with stakes, verifiable scores, and wallet-native flows.
Hackathons promise innovation—but their evaluation systems are often centralized, inconsistent, and non-transparent.
Despite building real, working products:
- Developers often receive subjective feedback due to limited judging bandwidth.
- It’s difficult to provide fully transparent and verifiable evaluation at scale.
- Self-assessment isn’t typically integrated into the process, missing an opportunity for deeper reflection and benchmarking.
- Results are usually event-bound, with limited long-term visibility or verifiability.
In contrast to transparent, high-stakes talent formats like India’s Got Latent—where participants own their confidence and results are visible—many hackathons still rely on opaque, off-chain decisions with no skin in the game.
How can we build a fully on-chain, AI-powered evaluation system for hackathons that is:
| Goal | Meaning |
|---|---|
| Decentralized | No sole reliance on a centralized judging authority for record-keeping and settlement |
| Verifiable | Scores, outcomes, and rewards recorded on-chain |
| Scalable | AI reduces judge bottlenecks using real project signals |
| Incentive-aligned | Participants stake value that reflects confidence (self-score vs eventual judge/AI score) |
- Developers connect wallets and submit projects (GitHub + live demo).
- They stake ETH via smart contracts (real on-chain transaction on Ethereum Sepolia).
- AI evaluates submissions from real URLs (repo + hosted site), not mocks.
- Scores are compared with self-assessment (1–10) in the product experience and results UI.
- Verified scores can be submitted on-chain via backend + agent flow; winners can claim through the contract where applicable.
- Wallet integration is required for user-signed txs
The UI, Graphify API, scoring API, and Sepolia contract are separate services.
The current contract address is configured in client/lib/contract.ts; use the
same address in both client and server environment files if you deploy another
instance.
- Copy
client/.env.exampletoclient/.env.local. - Copy
server/.env.exampletoserver/.envand setAGENT_PRIVATE_KEYto the private key for the trusted agent wallet. Keep this file private. - Start the graph API with
cd server && npm install && npm start. - Start the scoring API with
cd server/computer-use, create a Python 3.11+ virtual environment, installrequirements.txt, install Playwright Chromium, then runuvicorn main:app --host 0.0.0.0 --port 8000. - Start the client with
cd client && npm ci && npm run dev.
The owner must have already registered the agent wallet as trusted and created
an active round on Sepolia. To do that with the supplied helper, set
OWNER_PRIVATE_KEY to the contract owner and set AGENT_ADDRESS to the address
derived from AGENT_PRIVATE_KEY, then run node setup-contract.js from server.
The UI will switch the connected wallet to Sepolia before asking it to stake.
During evaluation, the browser sends only the entry ID. The server fetches its links and self-score from Sepolia, asks the Python scoring API to evaluate those values, and submits that score using the trusted agent wallet. This prevents a browser from choosing the score written on-chain.
For deployed setups, set NEXT_PUBLIC_SERVER_URL to the public Express API,
SCORING_API_URL to the Python scoring API, GRAPHIFY_API_URL to the Express
API, and CORS_ALLOW_ORIGINS to the exact client origins.