Lexicon Agent
With AI agents serve as partners in user's web3 journey. mass-adoption of blockchain is within reach!!
Created on 14th February 2025
•
Lexicon Agent
With AI agents serve as partners in user's web3 journey. mass-adoption of blockchain is within reach!!
The problem Lexicon Agent solves
Lexicon Agents: Your AI partner in web3 journey
Lexicon Agents support Lexicon Finance’s mission to make blockchain more accessible and user-friendly. Many users face challenges with transaction risks, investment decisions, and complex tools like Etherscan or Tenderly, which require technical expertise.
Lexicon Agents act as AI-powered partners, helping users securely manage their non-custodial accounts. They automate tool selection, generate easy-to-understand reports, and assist in decision-making, making blockchain interactions seamless and intuitive.
Three AI Agents on the Roadmap
🔹 Security Officer Agent (Alpha)
It contains 2 agents.
Risk Detector: Think of it as a virtual security analyst—it examines transactions just like a human expert would, identifying risks, gathering relevant data, and ensuring nothing gets overlooked.
Intent Matcher: Works like a personal transaction assistant, understanding user intent, cross-checking it with transaction details, and flagging any mismatches—just as a careful human would do before approving a transaction.
🔹 Financial Advisor Agent (Under Development)
Helps users make informed investment decisions by analyzing market trends and optimizing portfolios.
🔹 Intent Operator Agent (Under Development)
Automates blockchain actions by translating user intents into precise, executable transactions.
Challenges I ran into
Building fully autonomous AI agents requires them to identify tasks, gather necessary information, and determine when they have enough data to reach a conclusion. Another challenge is collecting, formatting, and labeling training data to ensure accuracy.
This challenge can be broken into two key questions:
1.What tools does a Web3 AI partner need?
2.How do we structure and prompt LLMs to perform tasks in a zero-shot manner?
For the first part, I solved it by analyzing the tools used by human professionals in blockchain security, finance, and operations. This helped define the necessary toolset for Web3 AI agents.
For the second part, since LLMs lack built-in decision-making for blockchain-specific tasks, we use a Zero-Shot ReAct (Reasoning + Acting) framework, allowing the model to:
1.Identify the task and retrieve necessary data.
2.Analyze transaction risks, intent mismatches, and financial insights logically.
3.Generate structured, human-like reports.
To guide LLMs effectively, we use structured prompts with:
Role definition (e.g., "You are a blockchain security analyst").
1.Step-by-step reasoning for logical task execution.
2.Integration with Web3 tools like risk scanners and transaction explorers.
3.By combining structured prompting, logical reasoning, and real-time data retrieval, Lexicon Agents enable LLMs to perform tasks autonomously, making blockchain safer and smarter.
Why This Order?
Training AI agents requires structured data, so we develop them in phases:
1.Security Officer Agent comes first, leveraging publicly available transaction and risk data for fine-tuning.
2.Financial Advisor Agent requires structured asset and strategy data, improving with intent-transaction mapping.
3.Intent Operator Agent is the most complex, as user intents vary widely. It depends on both the Security Officer and Financial Advisor for accurate execution.
By following this order, Lexicon Agents create a feedback loop, continuously improving AI agent’s accuracy.
Tracks Applied (7)