Sentra Overview
Sentra is M2Square's Risk Control Intelligence system for real-time payment risk scoring, analyst review, AI enrichment, and continuous model learning.
Sentra is the Risk Control Intelligence layer for M2Square payment infrastructure. It scores payment transactions in real time, explains the risk decision, routes escalated transactions to analysts, enriches KYC/KYB context asynchronously, and turns analyst feedback plus partner data into better risk models.
Sentra is not a static rules dashboard. It is an operating system for risk teams:
- A low-latency scoring API for payment authorization and routing decisions.
- A case review console for analysts and admins.
- A data contract for importing partner payment data, entities, labels, and KYC/KYB evidence.
- A local model operations loop for validating datasets, training XGBoost models on CPU, publishing artifacts to S3, and serving the active model in AWS.
- Optional LLM enrichment for KYC/KYB risk signals that never blocks the synchronous scoring response.
Operating planes
| Plane | What it does | Primary users |
|---|---|---|
| Local training plane | Validate contract-format datasets and train CPU XGBoost models. | Risk / ML / data team |
| AWS serving plane | Run sentra-api, load the active model pointer from S3, and serve /v1/risk/score. | Platform / backend team |
| Console plane | Provide analyst, admin, and demo workflows through the Next.js Sentra Console. | Analysts, admins, stakeholders |
Core runtime flow
sequenceDiagram
participant Platform as Payment platform
participant API as Sentra API
participant Model as Active XGBoost model
participant LLM as AI enrichment
participant Console as Sentra Console
Platform->>API: POST /v1/risk/score
API->>API: Upsert transaction + entities
API->>API: Build structured features
API->>Model: Score probability
Model-->>API: Fraud probability
API->>API: Map 0-1000 score to PASS/REVIEW/HOLD/BLOCK
API-->>Platform: Immediate risk decision
API-->>LLM: Queue KYC/KYB enrichment if cache miss
API->>Console: Persist risk score and case when escalated
LLM-->>API: Store AI signals and evidence
Console->>API: Analyst review + feedback labelsDecision actions
Sentra produces one of four decision actions:
| Decision | Meaning | Typical platform behavior |
|---|---|---|
PASS | Risk is below review threshold. | Continue payment flow. |
REVIEW | Transaction should enter analyst review. | Hold customer-facing fulfillment or route to manual queue. |
HOLD | Temporary freeze or delayed settlement is required. | Pause settlement, request verification, or wait for additional checks. |
BLOCK | Risk is above block threshold. | Stop the transaction and record the reason. |
Main product surfaces
| Surface | Path | Purpose |
|---|---|---|
| Cases | /cases | Analyst queue with filters, risk scores, decisions, and AI signal status. |
| Case detail | /cases/{case_id} | Transaction context, entities, latest score, evidence, feedback, notes, and timeline. |
| Manual Risk Check | /manual-score | JSON editor for calling /v1/risk/score and saving a score as a case. |
| Admin | /admin | Data contract, ingest runs, model registry, active model selection, and thresholds. |
Recommended implementation order
- Call Risk Scoring API from your payment decision point.
- Map your platform behavior to decisions and thresholds.
- Give analysts access to case review.
- Add KYC/KYB references for AI enrichment.
- Prepare historical data using the data contract.
- Build the feedback loop so reviews improve future models.