EAM

Ethereum Agent Manager (EAM) is an AI-powered tool that makes using Ethereum as easy as chatting. It understands natural language commands, like “Swap 1 ETH for USDC,” and turns them into secure blockchain transactions. With built-in security checks, EAM helps users avoid scams, risky contracts, and high gas fees. The system consists of a Flutter-based app (Charlotte) for smooth user interaction and an AI backend that processes requests and generates safe, executable blockchain actions. By simplifying Ethereum operations and ensuring security, EAM bridges the gap between Web2 and Web3, making blockchain accessible to everyone.

EAM is built using a combination of AI, blockchain, and modern frontend technologies to create an intuitive and secure Ethereum interaction layer. The frontend, Charlotte, is developed with Flutter, allowing it to support Linux, Android, and Web platforms with a single codebase. This ensures cross-platform compatibility and a seamless user experience. The AI backend is built using Python, leveraging Natural Language Processing (NLP) to understand user input and generate executable blockchain commands.

One of the core innovations is the Retrieval-Augmented Generation (RAG) system, which enables EAM to intelligently retrieve relevant blockchain commands and format them based on user intent. This system ensures accurate command matching while adapting to different Ethereum applications and protocols. The security check model is designed to analyze smart contracts and transactions in real-time, utilizing static analysis, behavior monitoring, and scam detection techniques to safeguard users from potential threats.

EAM follows a structured pipeline where user inputs are processed through Charlotte’s interface, passed to the AI backend for intent recognition and command generation, and then validated by the security check model before execution. The result is a natural language-driven, AI-powered Ethereum assistant that simplifies Web3 interactions while maintaining security and transparency.

A particularly hacky aspect of our implementation was optimizing the command generation and JSON formatting for blockchain transactions. Instead of relying on predefined templates, we designed a flexible formatting engine that dynamically adapts to different contract interactions. This allows EAM to support new protocols without requiring constant manual updates.