🔥 CloudTrain Studio

Personal Serverless MLOps Platform for Training, Governing, Versioning, Monitoring, and Serving Machine Learning Models.

---

🌟 Overview

CloudTrain Studio is an end-to-end, serverless Machine Learning Operations (MLOps) platform built entirely on the AWS Free Tier. It provides a beautiful, self-service dashboard to manage machine learning projects, trigger automated training pipelines, enforce model governance, and serve real-time predictions.

This project was built from the ground up to demonstrate production-grade Cloud Engineering, Event-Driven Architecture, and Modern Full-Stack Development.

Interview Pitch

"CloudTrain Studio is a personal serverless MLOps platform that allows me to create machine learning projects, upload datasets, automatically train models through event-driven AWS infrastructure, compare performance before promotion, track model history in Firebase, handle failures through SQS DLQs, and serve predictions through low-latency serverless APIs. The platform includes a modern React frontend, GitHub Actions CI/CD, observability, governance, and production-style architecture while remaining free-tier friendly."

✨ Features

• Event-Driven Training Pipeline: Uploading a CSV to S3 automatically triggers a serverless ML training pipeline using AWS Lambda and Scikit-Learn.
• Model Governance Engine: Automatically evaluates new models against the current production version. Promotes the model only if accuracy/R2 improves.
• Serverless Inference API: Blazing fast /predict REST API powered by API Gateway and Lambda, utilizing execution context caching to eliminate cold-start latency on repeated calls.
• Dead Letter Queue (DLQ): Robust error handling using Amazon SQS to capture failed training runs, bad data, or timeouts.
• Modern React Dashboard: A dark-mode first, glassmorphic UI built with Vite, React, Tailwind CSS v4, and Framer Motion.
• CI/CD Automation: Fully automated deployment to AWS and GitHub Pages via GitHub Actions.

🏗️ Architecture

The platform architecture follows best practices for scalable, serverless microservices:

1. Frontend: React (Vite) hosted on GitHub Pages.
2. API Gateway: REST API for serving predictions.
3. AWS Lambda (Inference): Caches models in memory and returns real-time predictions.
4. AWS Lambda (Training): Triggered by S3. Cleans data, trains models (Random Forest, Logistic Regression, etc.), and uploads artifacts.
5. Amazon S3: Stores raw CSV datasets and compiled .joblib models.
6. Amazon SQS (DLQ): Captures any failed events from the Training Lambda.
7. Firebase Firestore: Stores project metadata, training run history, metrics, and inference logs.

🚀 Deployment

The infrastructure is defined using AWS Serverless Application Model (SAM).

1. Configure the following GitHub Secrets in your repository:
   • AWS_ACCESS_KEY_ID
   • AWS_SECRET_ACCESS_KEY
   • AWS_REGION
   • FIREBASE_SERVICE_ACCOUNT (JSON string)
2. Push your code to the main branch.
3. GitHub Actions will automatically:
   • Build and deploy the AWS SAM stack.
   • Build and deploy the React frontend to GitHub Pages.

🛠️ Tech Stack

• Frontend: React, TypeScript, Tailwind CSS, Framer Motion, React Router.
• Backend: Python 3.12, Scikit-Learn, Pandas, Boto3, Pydantic, Firebase Admin.
• AWS: Lambda, API Gateway, S3, SQS, CloudWatch, IAM, SAM.
• Database: Firebase Firestore.

📈 Future Roadmap

• Model Drift Detection
• A/B Testing for deployed models
• Support for larger datasets via AWS Step Functions