LLM-based exploration and analysis of real-time and historical blockchain data

Blockchain technology has revolutionized digital transactions and decentralized applications through its transparent and immutable ledger system, with platforms like Ethereum processing millions of transactions daily. However, as blockchain networks grow, traditional blockchain explorers show limitations when providing intuitive access to this vast data landscape, particularly when handling complex analytical queries, interpreting transaction patterns, and serving users without technical expertise. In this paper, we address these limitations by proposing an intelligent blockchain explorer that combines a Large Language Model (LLM)-powered agent for real-time blockchain interactions with a schema-aware SQL agent for historical data analysis. For real-time interactions, a dedicated blockchain agent connects to live networks through external APIs and specialized tools to process queries about current transactions and network states. When analyzing historical data patterns, we use an approach in which a Retrieval-Augmented Generation (RAG) system enhances the SQL agent’s understanding of the blockchain database schema and structure. This SQL agent subsequently translates natural language queries into SQL commands for efficient data retrieval from our periodically synchronized blockchain database. A query processor, powered by an LLM, intelligently routes user queries between these components based on temporal and contextual requirements, which enables both immediate blockchain state analysis and complex historical data querying. We evaluate our system on diverse blockchain queries, including complex analytical scenarios and multi-step operations. The experimental results demonstrate the effectiveness of our schema-aware SQL agent in accurate query translation and the overall system’s capability in handling both real-time and historical blockchain data exploration tasks.