30+ algorithms across 4 families

See How Machine
Learning Actually Works

An interactive visualization platform powered by real scikit-learn computation. Explore decision boundaries, compare algorithms side-by-side, and watch models train step-by-step.

CONFLUENCE / LIVE
010+AlgorithmsREAL MODEL CATALOG
020DatasetsBUILT-IN GENERATORS
030FamiliesONE SHARED PIPELINE
040Boundary TypesGEOMETRIC TAXONOMY

Built for Understanding

Not another toy demo. Confluence gives you real algorithms with real computation, wrapped in an interface designed for exploration and comparison.

Scikit-Learn Boundary Overlay

Algorithm Knowledge Graph

Explore algorithms from first principles, mathematics, intuition, and visualization.

MACHINELEARNINGClassificationRegressionClusteringDim. ReductionLogistic Reg.SVMRandom ForestXGBoostNaive BayesKNNLinear Reg.RidgeLassoElastic NetGBRSVRK-MeansDBSCANGMMSpectralAgglomerativePCAt-SNEUMAPIsomap

Production-Grade Architecture

Follow an experiment from interactive configuration through real model computation, streamed training states, and high-performance visual rendering.

CONFLUENCE ENGINE
Intent → State

Configure Experiment

Dataset, algorithm, hyperparameters, and comparison slots become one synchronized experiment state that can be reproduced and shared.

Next.js · Zustand · URL state
01
Intent → State

Configure Experiment

Dataset, algorithm, hyperparameters, and comparison slots become one synchronized experiment state that can be reproduced and shared.

Next.js · Zustand · URL state
02
State → Contract

Validate Request

The client turns experiment state into typed payloads. FastAPI and Pydantic validate the contract before computation begins.

TypeScript · Axios · FastAPI · Pydantic
03
Contract → Model

Execute Real Models

Server-side scikit-learn estimators fit against generated datasets, preserving genuine model behavior instead of browser-side approximations.

Python · scikit-learn · SciPy
04
Model → Surface

Resolve Geometry

NumPy evaluates the fitted model over a prediction mesh and resolves probability grids, contours, metrics, and uncertainty data with deterministic cache keys.

NumPy · meshgrid · contours · cache
05
Iterations → Frames

Stream Evolution

Staged algorithms publish boosting rounds, tree depth, gradient descent, and neural epochs as WebSocket frames for immediate timeline playback.

WebSocket · async frames · staged training
06
Data → Understanding

Render Insight

The interface combines fast 2D heatmaps, precise SVG overlays, motion transitions, and optional 3D scenes into one interactive analytical surface.

Canvas2D · SVG · Framer Motion · Three.js
07
Services → Runtime

Run Consistently

Containerized services keep the Next.js interface, FastAPI computation service, and cache infrastructure reproducible across development environments.

Docker Compose · Next.js · FastAPI · Redis

Ready to Explore?

Open the visualizer, pick an algorithm, adjust hyperparameters, and see the decision boundary morph in real-time.

Launch Confluence