InVariants is the only no-code data intelligence platform that combines Topological Data Analysis (TDA), advanced dimensionality reduction, interactive clustering, time-series analysis and production-ready ML training - all from a single CSV file.
Powered by world-class open-source technologies
Most data tools - scatter plots, histograms, correlation matrices - are blind to the topology of your data: loops, voids, clusters, and connected components that encode the real story. InVariants brings algebraic topology to your fingertips with zero code required.
Upload a CSV and unlock a full analytical pipeline - from raw data to production model. Each module is fully interactive, with real-time Plotly visualizations and instant feedback.
A full no-code ETL pipeline. Impute missing values (mean, median, KNN), filter rows with AI natural language queries, one-hot encode categories, apply scalers (Standard, MinMax, Robust), detect and clip outliers (IQR, Z-score, Isolation Forest), and engineer new features - all with undo/redo history and recipe export.
Interactive auto-insights with group-by aggregation, full correlation matrix (Pearson / Spearman / Kendall), distribution plots with violin and box overlays, scatter matrices for relationship discovery, categorical bar charts, and association rule mining with configurable support and confidence thresholds.
Detect outliers and anomalies using three complementary approaches: Isolation Forest (tree-based, fast), Local Outlier Factor (density-based, novelty-enabled for new-data scoring), and topological kNN distance. Configure contamination rate, visualise anomaly score distributions interactively, save labels directly to your dataset, and export the fitted detector as a production-ready ZIP bundle with an inference script.
Six state-of-the-art algorithms with full parameter control, color-by-any-column, and one-click export of new dimensions back to the dataset for downstream analysis.
Six clustering algorithms with interactive parameter tuning, automatic optimal-k detection, silhouette score visualization, full dendrogram for Agglomerative, cluster profiling, and one-click label export as a new dataset column.
A complete toolbox for temporal data: from classical statistics to cutting-edge topological analysis of signals.
The heart of InVariants. Compute full persistent homology (H0, H1, H2), compare two datasets with Bottleneck and Wasserstein distances, generate persistence images and landscapes for ML feature extraction.
Build interactive KeplerMapper topological graphs with configurable cover parameters, DBSCAN clustering inside bins, and multiple lens projections (PCA, L2 norm, k-NN distance, eccentricity). Color nodes by any dataset column to reveal hidden cluster structure.
Extract TDA-derived numeric features (count, max persistence, entropy, amplitude per Hn) both globally and per-point using k-NN neighborhoods. Group-level TDA comparison by label column. Export as new dataset columns for ML input.
Train, evaluate and deploy classification and regression models with a single click. Explore model behaviour with Partial Dependence Plots and Individual Conditional Expectation (PDP+ICE). A full pipeline builder chains preprocessing (imputation, encoding, scaling) with the model for reproducible, exportable artifacts.
Topological Data Analysis captures the shape of data - holes, loops, connected components - that survive across scales. These are the features that distinguish normal operation from anomalies, one drug target from another, one network topology from a vulnerable one.
Full Vietoris-Rips complex computation via Ripser, the fastest persistent homology engine available.
Interactive Reeb-graph-style network visualization with 5 lens projections and real-time coloring.
Vectorized TDA representations ready to use as input to any ML model.
Track topological complexity along a signal - detect regime changes, anomalies, and periodicities invisible to FFT.
Formally compare two datasets using topology-based similarity metrics - not mere statistics.
Export H0/H1 features per observation using local k-NN neighborhoods - unique topological fingerprints for every data point.
From industrial IoT to financial fraud, any domain with complex, high-dimensional data benefits from topological insight.
Detect bearing failures and anomalous vibration patterns using sliding-window TDA on sensor streams. Phase portraits reveal attractor changes before classical thresholds trigger - hours earlier.
Apply TDA and Mapper to network flow data to distinguish normal traffic topologies from intrusion patterns. Topological features dramatically improve classifier performance on imbalanced attack datasets.
Transactions form distinctive topological clusters. Persistent homology reveals loop structures in fraudulent transaction networks invisible to standard correlation analysis and tree-based models alone.
Gene expression matrices and protein interaction networks have rich topological structure. InVariants enables unsupervised discovery of disease subtypes and drug response clusters that Euclidean methods miss.
Time-delay embeddings and Takens reconstruction uncover the attractor structure of financial time series. Regime changes appear as topological phase transitions detectable before volatility spikes.
Go beyond k-means. Mapper graphs reveal the true topology of your customer base - bridge customers connecting segments, isolated outlier cohorts, and natural lifecycle pathways invisible to cluster labels.
Flask server with session-based multi-user support, 100MB dataset handling, and a full undo/redo history engine.
The fastest persistent homology engine (Ripser) and the most expressive topological graph builder (KeplerMapper) - both integrated seamlessly.
Fully interactive, zoomable, downloadable charts - persistence diagrams, 3D scatter plots, Mapper networks, decomposition subplots and more.
Natural language filter generation: describe the rows you want in plain English and the platform generates the exact filter conditions automatically.
Every trained artifact is exportable as a production ZIP bundle: ML pipeline (scaler + model + predict.py), anomaly detector (Isolation Forest / LOF + detect.py), ARIMA model + forecast script, rolling anomaly config + script, and TDA sliding window config + inference script. Zero code needed.
HTTP Basic Auth, server-side session isolation, no data leaves your server. Deploy on your own infrastructure in minutes with systemd or Docker.
Drag and drop one or more CSV files. Smart auto-detection categorizes your columns instantly.
Impute, filter, encode, scale and engineer features using the visual pipeline builder.
Run TDA, dimensionality reduction, clustering, and time-series analysis with interactive charts.
Train ML models enriched with topological features, export pipelines, and run live inference.
Apply for beta access. We’ll review your request and send credentials directly to your inbox once approved — no credit card, no commitment.
No installation. No configuration. Upload a CSV and start discovering structure in minutes.