VaultNet

VaultNet is an AI-powered DeFi investment platform where users connect their MetaMask wallet via the React frontend, create customized trading vaults by specifying parameters like risk tolerance (1-10), trading style (conservative/moderate/aggressive), AI settings (sentiment influence, technical weights), and safety limits (stop-loss, take-profit), which triggers a database record creation in Hasura PostgreSQL followed by deployment of a Solidity smart contract to Sepolia testnet that holds the user's initial ETH deposit. Once created, a Node.js backend service runs a continuous trading loop every 5 minutes that fetches current ETH prices from Pyth Oracle, calculates technical indicators (RSI, MACD, SMA) via CoinGecko API, sends this market data along with vault configuration to a Python Flask server hosting a trained Deep Q-Network (DQN) neural network model (13 input features → 128→64 neurons → 3 Q-values for HOLD/BUY/SELL actions), receives AI predictions with confidence scores and position sizes, stores trading signals in the database, and executes trades by either swapping ETH to USDC on Uniswap (for SELL signals) or USDC back to ETH (for BUY signals) using the Vincent ability smart contract orchestration service with PKP wallet delegation for gasless transactions, all while users can monitor their vault performance, view AI trading signals with reasoning explanations, browse other vaults, invest additional funds, withdraw profits, and see real-time portfolio values through the dashboard that displays total value, profit/loss percentages, recent trades, and AI decision history pulled from GraphQL queries to Hasura, creating a fully autonomous AI trading system that combines machine learning predictions with blockchain execution and decentralized finance protocols.

VaultNet is an AI-powered decentralized finance (DeFi) investment platform that enables users to create and manage autonomous trading vaults powered by machine learning and blockchain technology. The platform integrates artificial intelligence, smart contracts, and real-time data analytics to make predictive trading decisions on behalf of users while maintaining full transparency and user control through Web3 technologies.

At its core, VaultNet operates as a fully decentralized trading ecosystem where users connect their MetaMask wallets through a React-based frontend. Once connected, they can create personalized trading vaults by configuring parameters such as risk tolerance (ranging from 1 to 10), trading style (conservative, moderate, or aggressive), and AI behavior settings that balance sentiment influence and technical analysis weight. Users also define safety mechanisms like stop-loss and take-profit thresholds to limit downside risks and lock in gains. When a vault configuration is submitted, the system creates a corresponding record in a Hasura-managed PostgreSQL database. This record stores all vault parameters and metadata. Simultaneously, a Solidity smart contract is deployed on the Sepolia testnet via the VaultFactory contract, which generates an individualized Vault.sol contract instance that securely holds the user’s initial ETH deposit.

Once the vault is deployed, VaultNet’s backend infrastructure — built in Node.js — takes over to manage the continuous trading lifecycle. A recurring task executes every five minutes, fetching real-time ETH price data from the Pyth Oracle for on-chain accuracy. It also retrieves complementary market information such as RSI, MACD, and SMA indicators from the CoinGecko API to form a complete technical snapshot of market conditions. This combined dataset, along with each vault’s configuration (risk level, sentiment weight, technical emphasis, etc.), is then sent to a Python Flask microservice. The Flask server hosts VaultNet’s proprietary Deep Q-Network (DQN) reinforcement learning model trained to identify optimal trading actions based on historical and live market data.

The DQN model consists of 13 input features representing both technical and sentiment-based metrics. It processes these inputs through two dense layers of 128 and 64 neurons, respectively, and outputs three Q-values representing possible actions: HOLD, BUY, or SELL. The action with the highest Q-value becomes the model’s decision for that trading interval, accompanied by a confidence score and an estimated position size (the proportion of the vault’s balance to trade). The Node.js backend receives this response and logs it into the Hasura database under the vault’s signal history, ensuring all AI decisions are auditable and transparent.

Based on the model’s action, VaultNet executes the necessary blockchain operations. If the DQN predicts a SELL signal, the system swaps a portion of the vault’s ETH for USDC via Uniswap, reducing exposure to potential price drops. Conversely, if the model predicts a BUY signal, it converts available USDC back into ETH to increase long exposure when bullish momentum is detected. For HOLD signals, the vault’s assets remain unchanged. All trading operations are executed securely and gaslessly using Vincent.