Project Overview

Astro AI Platform

A production-grade, enterprise AI platform for Telco Customer Churn prediction with full MLOps lifecycle.

Business Problem

Customer churn is one of the most critical challenges in the telecommunications industry. Losing a customer costs 5-25x more than retaining one. This platform leverages machine learning to predict which customers are likely to churn, enabling proactive retention strategies.

7,043
Customers
19
Features
Binary
Classification
~26%
Churn Rate
Data

Dataset Intelligence

Telco Customer Churn — features, preprocessing, and encoding strategy.

Categorical Features (15)

genderPartnerDependentsPhoneServiceMultipleLinesInternetServiceOnlineSecurityOnlineBackupDeviceProtectionTechSupportStreamingTVStreamingMoviesContractPaperlessBillingPaymentMethod

Numerical Features (4)

SeniorCitizenBinary (0/1)
tenure0-72 months
MonthlyCharges$18-$119
TotalChargesCumulative billing
Preprocessing
  • • Missing TotalCharges → Median imputation
  • • Categoricals → Label Encoding
  • • Numericals → Standard Scaling
  • • Train/Test split: 80/20 stratified
Models

Multi-Model Strategy

Six algorithms compete — the best is automatically crowned Champion.

Logistic Regression

Linear

Interpretable baseline

Random Forest

Ensemble

Robust to overfitting

Gradient Boosting

Boosting

Strong generalization

XGBoost

Boosting

Speed + performance

LightGBM

Boosting

Memory efficient

CatBoost

Boosting

Handles categoricals natively

Champion Selection Strategy

The Champion is selected using a balanced score that weighs multiple metrics:

20%
Accuracy
30%
F1 Score
30%
ROC-AUC
10%
Precision
10%
Recall

The model with the highest balanced score is automatically registered as Champion in the MLflow Model Registry and deployed for inference.

Technology

Enterprise Stack

Frontend

Next.js 15
TypeScript
Tailwind CSS
Framer Motion
Three.js
Recharts
React Query

Backend

FastAPI
Python 3.11
scikit-learn
XGBoost
LightGBM
CatBoost
Pydantic

MLOps

MLflow
DagsHub
Model Registry
Experiment Tracking
Artifact Logging

Deployment

Vercel
Render
Docker
Docker Compose
Health Checks
Deployment

Cloud Architecture

Frontend — Vercel

Next.js app deployed on Vercel's edge network with global CDN and automatic SSL.

Backend — Render

FastAPI backend deployed on Render with auto-scaling, health checks, and managed environment.

MLflow — DagsHub

Experiment tracking and model registry hosted on DagsHub with Git-integrated version control.