Maiat
The trust layer for the internet.
The problem Maiat solves
Every AI agent is about to start moving money, and we’re ignoring how easy it will be to manipulate them when they have no fucking clue who to trust.
In the near future, asking your AI to handle decisions will be normal:
“Find the best place to swap into ETH.”
“Book the highest-rated coffee shop nearby.”
“Buy this product at the best price.”
Your agent pulls from search results, review sites, and scraped data, all of which can be gamed with SEO tricks, fake reviews, botted engagement, and spam. Bad actors can very easily just manipulate what the agent sees, and the agent follows it blindly.
hence Maiat a trust layer agents can query before they act.
- You make a purchase on-chain, say you buy a coffee.
- Right after, Maiat messages you on Telegram: “How was it?”
- You leave a review, which is verified and is then added to Maiat database.
- The trust score then updates and the next person (or agent) asking where to buy gets an answer based on real experiences, not fake shit.
Challenges we ran into
The "5 chains, 1 weekend" problem
Integrating 5 different chain ecosystems (0G, KiteAI, Base, Hedera, Uniswap/Ethereum) in a hackathon is insane. Each has different SDKs, RPC quirks, and testnet faucet issues.
0G Compute — The 0G Serving Broker SDK had limited documentation for TeeML verification flows. We had to reverse-engineer the broker's expected request format by reading their Go source code. Ended up building a clean abstraction layer that handles provider selection and fallback gracefully.
Hedera HCS — Topic message submission kept failing silently. Turned out our transaction was hitting the max message size (1KB). Had to compress review attestation data and hash the full payload, storing only the hash + metadata on-chain with the full data in our DB.
Uniswap V4 Hook — Building a beforeSwap hook that queries an external oracle (our TrustScoreOracle) while staying within gas limits was tricky. We optimized by batching trust score updates off-chain and only storing the latest scores on-chain, so the hook does a simple mapping lookup instead of a cross-contract call during swap execution.
Chain mismatch bug — We accidentally deployed our on-chain review proofs to BSC Testnet instead of Base Sepolia. Caught it during end-to-end testing when the Base Verify integration couldn't find our contract. Quick fix but would've been embarrassing in a demo.
Use of AI tools and agents
In the product:
• 0G Compute Network — All review verification runs on 0G's decentralized AI inference. Models (Qwen 2.5, GPT-OSS, Gemma 3) analyze reviews for: content quality, spam detection, sentiment authenticity, and factual consistency. This runs inside 0G's Trusted Execution Environment (TeeML) so verification is verifiable and tamper-proof.
• Google Gemini — Powers the Telegram bot's conversational review flow. Users chat naturally with @MaiatBot, and Gemini extracts structured review data (rating, pros, cons, category) from free-form messages.
• Trust Score Engine — AI-computed composite scores combining: review sentiment analysis, on-chain activity correlation, reviewer reputation history, and cross-reference with market data (CoinGecko integration).
In the build process:• Claude Code (Anthropic) — Primary development agent. Built the entire Next.js frontend, all 5 chain integrations, and the smart contracts. Ran as autonomous background agents with specific tasks (one agent per integration).
• Codex (OpenAI) — Used for parallel code review and security auditing of smart contracts.
• AI-driven testing — Automated end-to-end verification of all bounty integrations using AI agents that simulated the full user flow.
The agents collaborated through a shared task board — one built features while another audited them. Human oversight on all merge decisions.
Shoutout- Jensen Huang
Tracks Applied (6)
Futurllama
Best Use of AI Inference or Fine Tuning (0G Compute)
0g Labs
Agent-Native Payments & Identity on Kite AI (x402-Powered)
Kite AI
Killer App for the Agentic Society (OpenClaw)
Hedera
Base Self-Sustaining Autonomous Agents
Base
Integrate the Uniswap API in your platform
Uniswap Foundation
Technologies used
