Enterprise-level Data Analysis AI Agent: An Intelligent System for Natural Language to SQL Conversion

This project proposes an enterprise-level intelligent data analysis system based on a multi-agent architecture, aiming to bridge the collaboration gap between business personnel and data engineers. The system supports natural language to SQL conversion, error correction, and schema awareness. Through the collaborative work of three core agents—planning, generation, and verification—it implements an intelligent data query solution for production environments.

Long-standing pain points exist in the field of enterprise data analysis: business personnel understand requirements but not SQL, while data engineers know SQL but struggle to grasp business context. This leads to simple queries requiring multi-party collaboration and low efficiency. This project provides an innovative solution to this disconnection problem.

The system is built on the LangGraph framework to form a multi-agent architecture, including three core agents:

1. Planning Agent: Parses user intent, decomposes tasks, analyzes dependencies, and selects strategies;
2. Generation Agent: Maps schemas, constructs SQL, adapts to dialects, and provides optimization suggestions;
3. Verification Agent: Checks syntax, verifies semantics, audits security, and estimates execution costs.

The system has two key technical features:

• Schema Awareness: Automatically learns database table relationships, field meanings, business term mappings, and data distributions, supporting fuzzy query understanding;
• Error Correction: Ensures SQL correctness through three layers: compile-time static analysis, runtime error repair, and result verification.

The project uses a modern technology stack:

• LangGraph: Multi-agent orchestration framework;
• LLM Backend: Supports OpenAI, Anthropic, and open-source models;
• Vector Database: Stores schema embeddings and term mappings;
• Database Connector: Provides unified access to mainstream relational databases. It also supports modular deployment, asynchronous processing, caching mechanisms, and audit logs to meet enterprise-level requirements.

Typical application scenarios include:

1. Self-service Data Analysis: Business personnel query directly using natural language;
2. Data Exploration and Discovery: Generate relevant queries through fuzzy questions;
3. Report Automation: Regularly generate templated reports. The project's value lies in promoting the evolution of enterprise AI applications from single-point tools to collaborative systems, better adapting to complex production environments, and providing a reference implementation for AI-driven data analysis.