AI SQL Assistant

Inspiration

Small businesses generate thousands of rows of valuable data every day—from sales records and inventory to employee information and customer details. However, most business owners don't know SQL and cannot afford expensive business intelligence tools or dedicated data analysts.

We wanted to remove this technical barrier by building an AI-powered assistant that lets anyone interact with their data using natural language. Our vision is simple: if you can ask a question, you should be able to understand your data.

What it does

AI SQL Assistant allows users to upload a CSV dataset and ask questions in plain English.

The application automatically imports the uploaded dataset into PostgreSQL, understands the database schema, generates optimized SQL queries using Google Gemini, executes them on the database, and displays accurate results instantly.

Instead of writing SQL manually, users simply upload, ask, and analyze their data through a conversational AI interface.

How we built it

We built the frontend using React, Vite, and Tailwind CSS to create a modern, clean, and responsive user experience.

The backend was developed using FastAPI and Python. Uploaded CSV files are processed using Pandas and stored dynamically in PostgreSQL using SQLAlchemy, allowing the application to work with any uploaded dataset without predefined database tables.

We integrated LangChain with Google Gemini as our primary Large Language Model for natural language understanding and SQL generation. Gemini analyzes the uploaded database schema, understands user intent, and generates optimized SQL queries from plain English questions. During development, we also leveraged Ollama with a locally hosted Large Language Model for rapid prototyping and offline testing.

The generated SQL is executed securely against PostgreSQL, and the results are returned to the frontend in an easy-to-read table. The complete application is deployed on Render, making it accessible from anywhere.

Challenges we ran into

One of our biggest challenges was making the application work with any uploaded CSV file instead of relying on predefined database tables.

We also faced several technical challenges including:

Dynamic schema detection
Prompt engineering for reliable SQL generation
Google Gemini integration with LangChain
PostgreSQL integration
Dynamic table creation
Frontend and backend communication
CORS configuration
AI response validation
Executing AI-generated SQL safely
Cloud deployment on Render

Solving these problems helped us build a much more flexible, scalable, and reliable AI-powered application.

Accomplishments that we're proud of

Built a complete AI-powered SQL Assistant from scratch.
Successfully integrated Google Gemini with LangChain for natural language to SQL generation.
Successfully converted natural language into executable SQL queries.
Enabled automatic CSV uploads into PostgreSQL.
Built a dynamic schema reader that works without hardcoded table structures.
Created an intuitive web application that requires zero SQL knowledge.
Successfully deployed both frontend and backend to the cloud.
Developed a solution that genuinely helps small businesses make data-driven decisions.

What we learned

This project taught us how to build an end-to-end AI application by combining frontend development, backend APIs, databases, cloud deployment, and Large Language Models.

We gained practical experience with: