🗄️ NL2SQL - Model Pipeline for Natural language to SQL QUERY

🚀 Overview

In many organizations, valuable data lives in relational databases, but accessing it requires SQL expertise. This project bridges that gap by allowing users to:

• Ask questions in natural language

• Upload ER diagram images

• Automatically generate validated SQL queries

All without writing SQL manually.

❓Problem Statement

In many organizations, data are stored in databases, but non-technical users can’t query it directly because they don’t know SQL. They rely on data analysts for even simple queries like.

Non-technical users struggle to query databases because:

• SQL has a steep learning curve

• Even simple queries require analyst intervention

• Complex queries take time and are error-prone

Example:

“Show me the total sales in 2024 by region.”

• Generating Effective and Complex for Non Technical Folk can be overwhelming
• Helps in generating complex query in seconds and saves time and cost for that specific organization

📖 Blog / Write-up

I’ve written a detailed blog that walks through the thinking, trade-offs, and occasional “why is this not working?” moments :) behind building this NL2SQL pipeline.

• How ER diagrams can be turned into executable schemas
• The design decisions that didn’t make it into the final code (for good reasons)
• And what actually breaks when theory meets production
  

Here's the link 👉

🧠 Solution

Build NL2SQL Pipline that helps users to quickly generation query regarding sql and also generate sql query by ER diagram We built an NL2SQL pipeline that converts:

• 📝 Natural language queries → SQL
• 🖼️ ER diagram images → Database schema → SQL And many more features...

User Input (Natural Language)
          ↓
   Text Preprocessing
          ↓
   Schema Understanding
          ↓
   NL2SQL Model (LLM or Encoder-Decoder)
          ↓
   SQL Query Generator
          ↓
   SQL Validator & Executor
          ↓
   Database (e.g., PostgreSQL/MySQL)
          ↓
   Results Display (Frontend/UI)

✨ Key Features

• 🔤 Natural Language → SQL generation
• 🖼️ ER Diagram image → Schema extraction → SQL
• 🧠 LLM-based reasoning with schema awareness
• 🔐 SQL security & validation checks
• 📋 Copy-ready SQL output
• 🧪 Syntax validation before execution

📊 Evaluation Metrics

• Syntactic Validit
  • Used syntactic Metrics that Observes if the query syntax is valid or invalid so if valid then query can be parseable
• Checks whether generated SQL is parseable
• Schema Consistency
  • Ensures correct table & column references
• Foreign Key Validation
• Security Rules
  • Blocks unsafe SQL patterns

Implementation

cd nl2sql
pip install requirements.txt

cd .\frontend
npm run dev

cd .\backend
uv run .\main.py

🧩 Example

📌 Future Improvements

• 🔄 SQL execution & result visualization
• 📊 Query optimization hints
• 🧪 Automated test coverage
• 🗂️ Multi-schema support
• 🔍 Reasoning and Streaming of Response

👩‍💻 Author

Aditya Katkar