A visual node-based architect for ML pipelines.
Designing ML pipelines is design work. It deserves a design surface — not a notebook, not a config file, not a graph of YAML files referencing one another across a repository.
RIVER gives ML engineers and researchers a visual node-based workspace for composing pipelines, verifying compatibility at every connection, and seeing the shape of the system before any training run begins.
When the architecture is right, RIVER writes the code.
A clean visual canvas for pipeline construction. Modular nodes for data, transforms, models, evaluation, and deployment.
Static type-checking of every connection — tensor shapes, dtypes, schema agreement — before training begins.
Trace data flow through the pipeline with synthetic batches. Find the broken edges in seconds, not hours.
Generate production-grade pipeline code (PyTorch, JAX, TensorFlow) from the visual graph — readable, idiomatic, version-controllable.
A growing library of standard ML primitives. Extend with your own nodes; share across the team.
Every pipeline is a versioned artifact. Branch, fork, compare, and roll back with confidence.
We give live walk-throughs to ML teams considering RIVER. No deck, no demo deck — just the application, against a problem of your choosing.