Web3 Action Pipeline

Seamless Translation of Natural Language to Blockchain Operations

The Web3 Action Pipeline describes the process of translating user inputs into on-chain actions and informative responses. Here's a detailed breakdown of each step:

Input Handling

• Frontend widget captures user query (e.g., "Swap 100 BERA for TRR")

• Input is sanitized and preprocessed for further analysis

Intent Classification

• LangGraph API determines if the request requires:

  • Pure information (RAG response)

  • On-chain action (tool execution)

  • Hybrid response (explanation + action)

• Classification is based on context, user history, and current market conditions

Parameter Extraction

• For action-based intents, key parameters are identified:

  • Token amounts

  • Addresses

  • Protocol-specific details (e.g., slippage tolerance, deadline)

• Natural language is parsed into structured data for smart contract interaction

Action Preparation

• For transactions:

  • Gas optimization calculations are performed

  • Wallet interaction flows are prepared (approvals, confirmations)

• For informational queries:

  • Relevant data sources are identified for retrieval

Execution

• On-chain actions:

  • Smart contract interactions are initiated

  • Transactions are signed and broadcast to the network

• Informational responses:

  • RAG system retrieves and synthesizes relevant data

  • LLM generates a coherent, context-aware response

State Management

• Maintains context across multi-step operations:

  `python*# Example: Compound yield strategy*

  1. Swap 50% of BERA for STONE

  2. Deposit STONE into liquidity pool

  3. Stake LP tokens in farm`

• Tracks progress of multi-step processes and handles interruptions or errors

Response Generation

• The LLM, via Bailoo, composes the final response

• Could be an informative explanation, an action confirmation, or both

• Tailored to user's expertise level and previous interactions

Logging & Monitoring

• Each step is logged in LangSmith for debugging and performance analysis

• Success rates, error patterns, and user satisfaction are tracked

• Data is used for continuous improvement of the system