Sentinel: Real-Time AI Fraud Detection

Enterprise-grade, event-driven anomaly detection pipeline with sub-millisecond ONNX inference.

---

Sentinel is an enterprise-grade, real-time fraud detection system. It simulates high-throughput financial transactions via streaming (Redpanda/Kafka) and evaluates them in milliseconds using an optimized ONNX inference engine.

Quick Start

Prerequisites

• Task (brew install go-task / choco install go-task)
• Docker & Docker Compose
• uv

Setup & Run

# 1. Clone repository
git clone https://github.com/enesgulerdev/sentinel.git
cd sentinel

# 2. Configure environment (Requires Google Drive File ID for gdown)
cp .env.example .env

# 3. Install dependencies via uv
task env:install

# 4. Execute ML Pipeline (Fetch dataset, preprocess, train baseline)
task ml:pipeline

# 5. Start microservices (API Gateway, Redpanda, etc.)
task docker:on

Container Management

task docker:on    # Start all services
task docker:down  # Stop gracefully (keeps images intact)
task docker:off   # Full wipe (removes containers, networks, volumes, images)

Local Services

Service	Local URL
API Gateway	http://localhost:8000
Redpanda	http://localhost:8080

Architecture & Deep Dives

Explore the sub-modules for advanced deployment, scaling, and observability patterns:

Module / Component	Description
Testing Suite	Unit, integration, and mock fixtures.
Helm Workloads	Autonomous local provisioning for stateful dependencies and isolated ML workloads.
GitOps & CD	Zero-touch deployment architecture using ArgoCD and Jenkins for deterministic state synchronization.
AWS FinOps Simulation	Infracost model demonstrating system scale to 14,500+ RPS with a 73% cost reduction under enterprise conditions.