interactive statistical machine learning

Try the whole ML process in your browser.

Not screenshots. Not code samples. Change the data assumptions, fit models, move thresholds, rerun validation, inspect uncertainty and watch the results change immediately.

Run the full workflow Browse 29 demos

29interactive demos

7workflow families

0server round trips

1live Flow/WASM app

Everything on this page runs locally. The statistical playground currently uses browser-native reference implementations; the MNIST classifier executes its numerical distance kernel in Flow-built WebAssembly. The new Flow → MLIR → wasm32 backend is being used to move the remaining workflows into the same runtime.

01 / complete workflow

Iris classification, end to end

interactive reference

Explore four measurements, optionally standardise them, create a held-out test set, fit K-nearest neighbours, predict every test sample and inspect the resulting accuracy and per-class recall.

02–28 / ML playground

Change one assumption. Rerun the process.

Every card includes data generation, fitting or statistical estimation, a live visualisation and quantitative diagnostics. Use the filters to jump directly to a part of the workflow.

29 / live Flow WebAssembly

Draw a digit and classify it.

Flow/WASM

Draw one handwritten digit. The browser normalises it to 28×28, then the compiled Flow kernel calculates its distance to ten learned MNIST prototypes. Nothing is sent to a server.

draw→crop→28×28→Flow WASM distance→class

Loading Flow WebAssembly and MNIST prototypes…

Predicted digit—Draw a digit, then classify it.

keep going

The browser is the front door, not the limit.

The repository contains the full Flow implementations, benchmark suite, tests and examples behind the project. The browser playground is being progressively switched from reference JavaScript to the direct MLIR/WASM target as the runtime surface becomes available.

Browse the library ↗ Run the Flow examples ↗ Inspect benchmarks ↗ Inspect tests ↗