plexe-ai/plexe — reverse-engineered prompt

Build me a Python tool called Plexe that lets me create a machine learning model just by describing what I want in normal language and pointing it at a tabular dataset like CSV, Parquet, ORC, or Avro. I want to run it from the command line or from Python, give it an intent like “predict customer churn,” and have it automatically inspect the data, figure out the task and metric, try a few good model options, evaluate them, and save the best result as a deployable model package with inference code, schemas, config, metrics, and a README.

Please make it focused on tabular machine learning, with support for common model families like XGBoost, CatBoost, LightGBM, Keras, and PyTorch where appropriate. I also want a simple YAML config for things like search iterations, model restrictions, and LLM routing, plus a small dashboard to inspect experiment results from a work directory. It should work with OpenAI and Anthropic style API keys, and Docker support would be great for an easy setup. Look up current docs online if you need to.