Skip to main content
ML-Powered Adaptive Routing

Notifications That Learn Where Your Users Listen

NotifyEngine uses an XGBoost ML model to predict the best delivery channel for every recipient - email, SMS, push, or webhook - reaching users on the channels where they actually listen. No rules to configure. No workflows to build. It just learns.

Explore the Docs View on GitHub
See how it works Explore features
4
Channels
15
Features Extracted
Real-time
Adaptive
ML Channel Router
XGBoost v2.1
incoming notification
recipient: user_abc
template: order-shipped
Email 82% ✓
Push 45%
SMS 31%
Webhook 18%
optimal route selected
Email (82% predicted engagement)
The Problem

Every Notification Platform Treats Every User the Same

Today's notification tools give you two choices: raw infrastructure where you build everything yourself, or workflow platforms where you hard-code static rules that never adapt. Neither learns. Neither optimizes. Neither gets smarter over time.

Compare all platforms
Infrastructure

"Here's a pipe, figure it out."

AWS SNS, Twilio, SES - you call an API, it sends. No routing logic, no intelligence, no optimization. You build everything yourself.

Workflow Platforms

"Build your own rules, we'll follow them."

Courier, Knock, Novu - drag-and-drop workflow builders where a human hard-codes channel priority. Same rules for every user, every time.

NotifyEngine
Adaptive Intelligence

"We learn what works and route automatically."

NotifyEngine uses ML to predict the best channel per user, per moment. No human in the loop. The system continuously adapts as behavior changes.

Market Position

Every Competitor Lets You Build Workflows. None of Them Learn.

The notification market has three tiers. NotifyEngine is alone in the third - the intelligence layer that nobody else occupies.

Infrastructure
Dumb Pipes
AWS SNS/SES Twilio
Multi-channel delivery
Routing logic
Workflow builder
Per-user optimization
ML-powered routing
Self-improving
Workflow Platforms
Rule-Based
Courier Knock Novu
Multi-channel delivery
Routing logic
Workflow builder
Per-user optimization
ML-powered routing
Self-improving
Adaptive Intelligence
ML-Powered
NotifyEngine
Multi-channel delivery
Routing logic
Workflow builder
AUTO
Per-user optimization
ML-powered routing
Self-improving

Amazon Pinpoint - the closest enterprise attempt at intelligent notifications - is shutting down October 2026.

How It Works

From API Call to Intelligent Delivery

Every notification passes through five stages. The entire flow - from request to delivery to learning - happens in milliseconds.

Explore all features
  1. STEP 01

    API Request

    Your application sends a notification via a simple REST API call. Specify the recipient, template, and payload - nothing else needed.

    POST /v1/notifications
  2. STEP 02

    Feature Extraction

    The system extracts 15 behavioral features per channel: time of day, day of week, historical engagement rate, notification fatigue signals, channel health, and more.

    15 features × 4 channels
  3. STEP 03

    XGBoost Prediction

    The ML model scores each available channel with an engagement probability. Trained on rolling 30-day delivery history for each individual recipient.

    Email: 82% · Push: 45% · SMS: 31%
  4. STEP 04

    Epsilon-Greedy Exploration

    90% of the time, route to the highest-scoring channel. 10% of the time, explore alternatives to discover behavioral changes and prevent stale predictions.

    ε = 0.10 exploration rate
  5. STEP 05

    Delivery + Feedback Loop

    Notification delivered via the optimal channel. Engagement tracked through tracking pixels, WebSocket receipts, and webhook confirmations - feeding back into the model.

    Continuous learning cycle
Features

Everything You Need. Nothing You Don't.

Production-grade notification infrastructure with ML intelligence.

See the architecture

  • Adaptive Channel Routing

    XGBoost ML model predicts the optimal delivery channel per user in real-time. 15 behavioral features, rolling 30-day training window.

  • Multi-Tenant Architecture

    Full tenant isolation with API key authentication, per-tenant rate limiting, usage metering, and scoped data boundaries.

  • Multi-Channel Delivery

    Email, SMS (webhook), in-app (WebSocket), and generic webhook - all from a single API call. The ML model picks the best one.

  • Intelligent Retry & Recovery

    Priority-based exponential backoff with jitter, dead letter queues for failed deliveries, and circuit breakers per channel provider.

  • Real-Time Engagement Tracking

    Tracking pixels for email opens, WebSocket read receipts for in-app, webhook confirmations - automatic feedback loop to the ML model.

  • Full Observability

    Prometheus metrics, Grafana dashboards, Pino structured logging, and per-tenant analytics. Know exactly what's happening.

  • Continuous Learning

    Epsilon-greedy exploration ensures the model adapts as user behavior evolves. 10% exploration rate prevents stale predictions.

  • Developer-First API

    Clean RESTful endpoints, idempotency keys, versioned API paths, comprehensive error responses, and webhook event subscriptions.

Developer Experience

One API Call. ML Does the Rest.

Send a notification with a single POST request. NotifyEngine handles channel selection, delivery, retries, and learning - automatically.

How the architecture works
send-notification.ts 
// Send a notification - NotifyEngine handles the rest
const notification = await fetch('https://api.notifyengine.io/v1/notifications', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer ne_live_abc123',
    'Content-Type': 'application/json',
    'Idempotency-Key': 'order-shipped-12345'
  },
  body: JSON.stringify({
    recipient: 'user@example.com',
    template: 'order-shipped',
    priority: 'high',
    payload: {
      orderId: '12345',
      trackingUrl: 'https://track.example.com/12345'
    }
  })
});
response.json
200 OK 
// Response includes ML routing decision
{
  "id": "ntf_7x8k9m2n",
  "status": "queued",
  "routing": {
    "selected_channel": "email",
    "confidence": 0.82,
    "reason": "XGBoost predicted highest engagement for email",
    "alternatives": {
      "websocket": 0.45,
      "sms_webhook": 0.31
    }
  }
}

ML routed to email with 82% predicted engagement

Architecture

Built for Scale and Intelligence

A production-grade event-driven architecture with the ML service at its core - not as an afterthought.

View the full tech stack