Your risk desk
What this account has scanned and scored, and where to pick up next.
Find the fraud buried in the file
Upload a transaction CSV and the platform maps the columns, derives twelve behavioural features and scores every row through a Random Forest and gradient-boosted ensemble. The queue that comes back is ordered by expected loss, and each case names the features that put it there.
One score, and the reasons behind it
Type the figures in or upload a file of companies. You get a health score out of 100 built from four transparent pillars, a model estimate of failure probability over 24 months, the drivers behind it, and recommendations that carry the number closing each gap.
How the platform decides, and what it scored
Seven methods, each with the live evidence that it ran in this session — real hyperparameters, real held-out metrics, and an auditable log of every fit and every scoring batch. The models below are not descriptions of models. They are the models.
Your recent activity
Every sign-in, scan and health score this account has run, newest first.
Built as a self-contained browser application: no server, no database, and no external machine-learning library. The tree ensembles, the logistic baseline, the transformer encoder and the SHAP and LIME explainers are implemented in assets/ml.js and execute locally every time the platform scores. Reference datasets are synthetic and generated in-page from documented processes, so nothing you load leaves the browser.