Back to home
Full platform capability map

Features

Every capability in FraudVerse AI, what it does, and exactly how to use it — from raw transaction data to an explainable, actionable fraud decision.

XGBoostFastAPIPandas

ML Fraud Scoring Engine

Real-time transaction risk scoring

An XGBoost classifier trained on transactional and behavioral features scores every account/transaction for fraud probability, validated at 89.2% recall and ROC-AUC 0.9737.

How to use

  1. 1Go to Predictions from the sidebar.
  2. 2Submit a feature payload (JSON) or upload a CSV of accounts.
  3. 3Click Run Prediction to get a fraud probability score instantly.
SHAPTreeExplainer

SHAP Explainability

See exactly why an account was flagged

Every prediction is decomposed feature-by-feature using SHAP (TreeExplainer), so investigators get a transparent breakdown instead of a black-box number.

How to use

  1. 1After running a prediction, click Explain This Prediction.
  2. 2Review the top contributing features and their SHAP impact.
  3. 3Use this breakdown to justify investigator decisions.
NetworkXCommunity Detection

Graph Intelligence

Uncover mule networks & fraud rings

Accounts are modeled as a network graph. Community-detection algorithms surface clusters of connected accounts that isolated, single-account scoring would miss entirely.

How to use

  1. 1Open Graph Intelligence from the sidebar.
  2. 2Click any node to inspect its risk profile and connections.
  3. 3Toggle 'High risk only' to focus on flagged clusters.
Policy Engine

Tiered Response System

Risk scores → concrete actions

Every risk score maps to a policy tier: low (silent log), medium (soft notify + user confirms), or high (hard hold + step-up verification + investigator alert).

How to use

  1. 1Tiers are computed automatically wherever a risk score appears.
  2. 2Check /api/v1/tiers/thresholds for the documented cutoffs.
  3. 3High-tier accounts surface directly in the Investigation queue.
Client-side CSV Parsing

Dataset Upload

Bring your own transaction data

Drag-and-drop a transaction CSV for client-side preview and validation before it's sent for scoring — nothing leaves your browser until you submit.

How to use

  1. 1Go to Upload from the sidebar.
  2. 2Drag a UTF-8 encoded .csv (up to 25MB) or click to browse.
  3. 3Review the parsed preview, then proceed to scoring.
Next.jsRecharts

Live Fraud Dashboard

Real-time visibility at a glance

A unified view of fraud cases, high-risk accounts, cluster activity, business impact, and a live simulated transaction feed — all backed by real pipeline endpoints.

How to use

  1. 1Dashboard is the default landing page after login.
  2. 2Watch the Live Transaction Feed for real-time scoring events.
  3. 3Use Graph Summary + Investigation Summary cards to jump to details.
Gemini AIRegistry API

Customer Alert Center

Plain-language alerts for end users

A customer-facing view that explains flagged transactions in simple, non-technical language, plus a pre-payment Receiver Trust Score check and crowdsourced fraud registry.

How to use

  1. 1Open Customer Alerts from the sidebar.
  2. 2Review flagged/resolved transactions and their explanations.
  3. 3Use 'Check a Payee Before Paying' before sending money.
FastAPIAudit Logging

Compliance Reporting (STR)

Regulator-ready suspicious transaction reports

Generate Suspicious Transaction Reports for any account on demand, plus exportable cluster-risk and model-status summaries for audit trails.

How to use

  1. 1Go to Reports from the sidebar.
  2. 2Enter an Account ID under Compliance Report (STR).
  3. 3Click Generate to produce a regulator-ready report.
Gemini 2.5FastAPI Proxy

AI Assistant (Gemini)

Ask anything about fraud or the platform

A floating in-app assistant, powered by Gemini, answers questions about how to use FraudVerse AI and general banking/fraud/KYC concepts in clear, point-wise, easy-to-remember answers.

How to use

  1. 1Click the floating chat icon in the bottom-right corner, on any page.
  2. 2Ask a platform question ('how do I generate an STR report?') or a concept question ('what is a mule account?').
  3. 3Answers are formatted point-wise with key terms highlighted for quick scanning.
Server-Sent Streaming

Live Streaming Simulation

Watch scoring happen in real time

A simulated real-time transaction stream lets you observe the ML engine scoring accounts live — useful for demos and for stress-testing the pipeline end-to-end.

How to use

  1. 1The Live Transaction Feed on Dashboard runs automatically.
  2. 2Each event shows the account, fraud probability, and flag status as it's scored.
Audit LoggingField Tokenization

PII Tokenization & Security

Bank-grade data protection by default

Sensitive fields are tokenized at rest before storage, and every account access is captured in an audit log — who looked at what, and when.

How to use

  1. 1Enabled automatically on the backend — no user action needed.
  2. 2Set PII_TOKEN_SECRET in your environment before any real deployment.