NLP to SQL Queries

Plain english to SQL queries to interact with database deployed on AWS using EKS.

GitHub Repository: View on GitHub

Introduction

This project demonstrates how to build and deploy a FastAPI application that translates natural language questions into SQL queries and runs them against a dataset hosted in Amazon Athena. The entire pipeline is containerized, stored in Amazon ECR, and deployed on an AWS EKS cluster, accessible via a public endpoint.

Problem

Data analysts and non-technical stakeholders often struggle to query data if they don’t know SQL. A natural language interface can bridge this gap, allowing anyone to get insights from data using plain English. In this service, users can type a question in plain English (e.g., “Show me all products ordered in July”), and the system generates the corresponding SQL query, executes it against the database, and returns the results in a CSV file - all without touching SQL.

Approach

  1. Application Development
    • Built with FastAPI to handle HTTP requests.
    • Integrated OpenAI API to translate natural language into SQL queries.
    • Connected to Amazon Athena to execute SQL queries on S3-hosted data.
  2. Containerization
    • Created a multi-architecture Docker image supporting amd64 and arm64.
    • Stored images in Amazon ECR.
  3. Continuous Integration & Deployment
    • Used GitHub Actions to automate:
      • Building the Docker image.
      • Pushing it to ECR.
      • Deploying updates to EKS.
  4. Infrastructure
    • Deployed on an AWS EKS cluster.
    • Exposed via a LoadBalancer service to get a public URL.

Results

  • Successfully deployed a public API where users can send natural language questions and receive SQL results.
  • Achieved end-to-end automation from code commit to deployment.
  • Validated scalability on AWS infrastructure.

Challenges

  • GitHub Actions not triggering due to missing workflow events.
  • Docker architecture mismatch (built on ARM Mac, deployed on AMD AWS nodes).
  • OpenAI API quota issues causing runtime errors.
  • Kubernetes CrashLoopBackOff errors due to missing environment variables.
  • Cost management — keeping EKS running racks up charges quickly.

How I Solved

  • Added on: push triggers in GitHub Actions to automate builds.
  • Used --platform linux/amd64 in Docker builds to match AWS node architecture.
  • Stored sensitive credentials (OpenAI API key, AWS creds) in GitHub Secrets and Kubernetes secrets.
  • Implemented cleanup scripts to delete EKS clusters, ECR repos, and S3 data to avoid extra costs.

Conclusion

This project demonstrates the full lifecycle of deploying an NLP-to-SQL application in the cloud, from local development to production deployment on AWS. It also highlights the importance of CI/CD, cross-platform containerization, and cost management.

Future Improvements

  • Implement caching to reduce repeated OpenAI calls and save on API costs.
  • Add authentication and rate-limiting for the public API.
  • Expand support for multiple databases beyond Athena.
  • Enhance natural language processing for more complex queries.
  • Add monitoring and logging dashboards using AWS CloudWatch and Grafana.