Overview
What is it?
Kolosal AutoML is a single-crate Rust AutoML framework: no Python runtime, no autodiff dependency — the training engines, hyperparameter search, and neural architecture search are all hand-implemented over ndarray. It covers the full pipeline from raw tabular data to a served, explainable model in one binary.
You get a dataset in, the platform searches over 17+ algorithms and tunes them with a Bayesian/TPE optimizer (HyperOptX), then hands back a calibrated, drift-monitored model you can query through a REST API or an Axum-served htmx dashboard.
Features
What it does
- Multi-Model AutoML Search — 17+ training algorithms plus Bayesian/TPE/Hyperband hyperparameter search, auto-assembled into a full pipeline.
- Model Trust Tooling — Explainability (permutation importance, partial dependence, SHAP-like local explanations), probability calibration, drift detection, and anomaly detection.
- ISO Compliance Layer — Fairness/bias scanning, PII detection, and data-lineage tracking, working toward ISO 42001/27701/TR 24027 readiness.
- Axum Web Dashboard & CLI — Upload, configure, train, and monitor runs from a browser dashboard or a
train/predict/serveCLI, sharing one binary. - Inference & Export — A batching/caching inference engine with ONNX/PMML export and model versioning.
- Fully Open Source — No vendor lock-in. Deploy anywhere, extend freely, and contribute back to the community.