Looking for a job

sachin masti

@sachinmasti

Machine Learning Engineer who shipped three production ML systems, from spam detection to NYC Airbnb pricing and churn prediction.

India

Message

I independently shipped three production machine learning systems: a spam classifier, an NYC Airbnb pricing model, and a customer churn predictor. I trained, tuned, and deployed each end-to-end.

For my spam detection system, I used TF-IDF and an Optuna-tuned LightGBM classifier, then addressed class imbalance with SMOTE. It reached 99.63% test AUC and was shipped as a real-time FastAPI endpoint in Docker.

For the Airbnb NYC Price Predictor, I cleaned listing data and engineered pricing signals, including Times Square proximity and keywords in listing titles. I tuned a CatBoost regressor and deployed the app on Render via Flask, with a fallback estimator.

I also built ChurnGuard AI, using a scikit-learn pipeline and an ensemble of XGBoost classifiers. I deployed it as containerized FastAPI and Gradio services with Docker Compose on Render, including a dashboard for model metrics and feature distributions.

Experience

Work history, roles, and key accomplishments

Airbnb NYC Price Predictor

GitHub

Cleaned a 48,895-listing NYC Airbnb dataset and engineered pricing signals including Times Square proximity and TF-IDF keyword detection. Tuned a CatBoost regressor to R² = 0.63 and deployed on Render via Flask with a fallback estimator.

CatBoost Feature Engineering TF IDF Flask Deployment Data Cleaning

Customer Churn Prediction System

ChurnGuard AI

Built an end-to-end churn prediction system on a ~20,000-customer dataset with a scikit-learn pipeline and a bagging ensemble of 150 XGBoost classifiers. Deployed as two containerized services via Docker Compose on Render with a live insights dashboard.

XGBoost scikit learn Ensemble Learning Docker Compose fastAPI Gradio Deployment

Spam Detection System

GitHub

Built an end-to-end spam classifier using TF-IDF vectorization and an Optuna-tuned LightGBM classifier on a 5,168-message dataset. Achieved 99.63% test AUC and 96.8% precision on the spam class, and deployed as a real-time FastAPI endpoint in Docker.

TF IDF LightGBM Optuna SMOTE Cross Validation fastAPI Docker Pydantic