Mrityunjay Pathak

Developed a churn prediction system on 7,000+ customer records from PostgreSQL using Scikit-learn pipelines, ensuring reproducible training and preventing data leakage.

Benchmarked seven classification models through MLflow experiment tracking, selecting Logistic Regression for matching PR-AUC and recall with greater interpretability.

Tuned the decision threshold via the precision-recall curve, raising recall from 80% to 90% while accepting a precision drop from 49% to 43% as missing a churner outweighs a false retention offer.

Integrated a Dockerized FastAPI backend with a React frontend, pulling the trained model from Hugging Face Hub as a remote artifact store for on-demand risk scoring.

Built an ML pipeline to predict used-car prices on 2,800+ vehicle listings scraped from Cars24 via Selenium and BeautifulSoup.

Stacked XGBoost, Random Forest, and Gradient Boosting into a tuned ensemble, cutting MAE by 31% (₹1.23L to ₹85K) and improving R2 from 0.77 to 0.88 over a Linear Regression baseline.

Containerized the model with Docker and exposed it as a FastAPI service on Render, serving real-time price predictions through a frontend application.

Designed an ETL pipeline with Python and SQLAlchemy to load 50,000+ sales records into a Neon PostgreSQL database, powering downstream reporting and analysis.

Automated the workflow using GitHub Actions to ingest a simulated daily feed of ~100 new transactions, achieving zero failures across 250+ runs at ~47 seconds each.

Delivered an interactive Power BI dashboard with scheduled refreshes to track sales performance, revealing Q4 as the strongest quarter at 27% of annual revenue.