Prism

Prism is an end-to-end machine learning pipeline for tabular data, taking a dataset from raw file to served model. It loads CSV, JSON, or Parquet with schema validation, then handles preprocessing (missing values, encoding, normalization, and skew correction) before generating interaction and polynomial features and running feature selection. Training covers RandomForest, GradientBoosting, LogisticRegression, SVM, and ElasticNet, with cross-validation, hyperparameter search, and MLflow-compatible experiment tracking. Trained models are served behind a FastAPI endpoint. The whole flow is driven from YAML config through a three-command CLI: train, evaluate, and serve.

Technologies Used

Python

scikit-learn

FastAPI

Pandas

NumPy

Docker