plexe

Why we included this project

Most teams building a predictive model spend their time on trial and error, rewriting training scripts and tuning parameters. Plexe takes a plain-English description of the problem plus a tabular dataset in CSV, Parquet, ORC, or Avro and runs that loop for you. Its agents figure out the task type and the right evaluation metric, then iterate through candidate models from XGBoost, CatBoost, LightGBM, Keras, or PyTorch and return the best performer with a report. The run is split across a six-phase workflow, so you get a documented, reasoned result rather than a black-box grid search. You can drive it from a single CLI command or call it from Python, which keeps experiments scriptable and easy to control.