Technology Stack
Extraction normalization reconciliation and alerts
Ingest and normalization
Secure intake of PG and bank settlement events and ERP and POS snapshots. Store in versioned object storage and apply schema checks.
Features and models
Field extraction from documents and exports
Time series anomalies for sales and refunds
Hybrid policy rules and gradient boosted models to produce riskScore and riskGrade
Inference
Online checks target low latency while batch windows handle large reconciliation jobs.
Explainability and MLOps
Reason codes and top drivers for each alert
Drift checks snapshots and rollback with human in the loop
Output to RVA
Scores grades and reasons feed the RVA builder and may later be referenced by partners under their own frameworks.