Skip to content

Repository files navigation

Sentinel: Real-Time AI Fraud Detection

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

Go Version Python Version Rust Version Latest Release License


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.

Performance Benchmark Peak Performance Benchmark: Sustaining 25,300+ RPS with 7.79ms average latency over 4.5 million requests.

Note: Note: Peak 25,300+ RPS was achieved under optimal hardware conditions. Standard local Docker Desktop deployments typically yield ~18,000+ RPS due to local CPU and network bridge constraints.

graph TD
    %% Styling
    classDef go fill:#00ADD8,stroke:#fff,stroke-width:2px,color:#fff;
    classDef rust fill:#DEA584,stroke:#fff,stroke-width:2px,color:#000;
    classDef python fill:#FFD43B,stroke:#306998,stroke-width:2px,color:#306998;
    classDef infra fill:#f9f9f9,stroke:#333,stroke-width:2px;
    classDef storage fill:#ff9900,stroke:#fff,stroke-width:2px,color:#fff;
    classDef broker fill:#8B0000,stroke:#fff,stroke-width:2px,color:#fff;

    %% Nodes
    Client([Client / K6 Load Tester])

    subgraph API Layer
        Gateway[Go Gin Gateway]:::go
    end

    subgraph Streaming & Validation
        Redis[(Redis<br>Idempotency)]:::infra
        Broker1{Redpanda<br>raw-events}:::broker
        Validator[Rust Stream Processor]:::rust
        Broker2{Redpanda<br>clean-events}:::broker
    end

    subgraph AI & Persistence
        Consumer[Python Inference Engine<br>ONNX Model]:::python
        S3[(AWS S3 / LocalStack<br>Audit Logs)]:::storage
        DB[(PostgreSQL)]:::infra
    end

    %% Edges (Flow)
    Client -->|HTTP POST| Gateway
    Gateway -->|1. Check tx_hash| Redis
    Redis -.->|Duplicate? Block| Gateway
    Gateway -->|2. Fire & Forget| S3
    Gateway -->|3. Publish| Broker1
    Broker1 -->|4. Consume Batch| Validator
    Validator -->|5. Type Check & Validate| Validator
    Validator -->|6. Publish Validated| Broker2
    Broker2 -->|7. Consume Batch| Consumer
    Consumer -->|8. Fraud Inference| Consumer
    Consumer -->|9. Persist Result| DB
Loading

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

# 6. Run load tests to verify system performance
k6 run tests/fixtures/loadtest.js

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 25,300+ RPS with an 85% cost reduction under enterprise conditions.
Policy & Governance Enterprise Policy-as-Code standards enforcing infrastructure, container, and Kubernetes security via OPA/Rego.
AI Release Agent Autonomous, AI-driven Python agent for dynamic Semantic Versioning and automated release notes generation via Gemini.
AI Doc Agent Autonomous, AI-driven Engineering Council that analyzes git diffs to generate weekly persona-based architectural reviews.
Local AWS Simulation Fully offline AWS S3 audit logging and event routing simulation via LocalStack.

About

Sub-millisecond AI fraud detection at enterprise scale. Engineered for high-throughput predictability.

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages