System Objective

1. System Objective

Design a real-time, scalable fraud prevention platform that detects suspicious blockchain transaction behavior before settlement by combining:

• Event streaming and real-time feature engineering (Confluent)

• Machine-learning–based risk scoring (Google Cloud Vertex AI)

• Policy-driven alerting and intervention (off-chain)

---

2. Architectural Principles

• Event-driven (no batch dependencies)

• Low latency (sub-second detection)

• Stateless services

• Cloud-native & managed

• Explainable AI decisions

• Blockchain-aware but chain-agnostic

---

3. High-Level Architecture Overview

┌────────────────────┐
│ Blockchain Network │
│ (Node / Indexer)   │
└─────────┬──────────┘
          │ Transactions
          ▼
┌──────────────────────────┐
│ Confluent Cloud (Kafka)  │
│  - raw-transactions      │
│  - enriched-transactions │
│  - fraud-alerts          │
└─────────┬────────────────┘
          │ Feature Events
          ▼
┌──────────────────────────┐
│ Stream Processing Layer  │
│ (ksqlDB / Kafka Streams) │
│ - Windowed features      │
│ - Behavioral metrics     │
└─────────┬────────────────┘
          │ Feature Vector
          ▼
┌──────────────────────────┐
│ Google Cloud Run         │
│ Fraud Scoring API        │
└─────────┬────────────────┘
          │ Inference Call
          ▼
┌──────────────────────────┐
│ Vertex AI                │
│ Fraud Detection Model    │
└─────────┬────────────────┘
          │ Risk Score
          ▼
┌──────────────────────────┐
│ Decision & Policy Layer  │
│ (Cloud Run)              │
└─────────┬────────────────┘
          │ Alerts / Logs
          ▼
┌──────────────┬───────────────┐
│ Fraud Alerts │ BigQuery       │
│ Dashboard    │ Audit / Train  │
└──────────────┴───────────────┘