Results at a Glance
99.9%
System uptime on AWS, sustained in production
<1s
Query response time across millions of asset events
Millions
IoT events ingested and processed daily in real time
AI-ready
Predictive maintenance and anomaly detection models running on live data
The Problem
A construction technology company managing a large fleet of heavy assets needed a backend capable of ingesting millions of IoT sensor events per day, including GPS pings, utilisation metrics, engine telemetry and geofence triggers, and serving that data to real-time dashboards used by site managers and operations teams across multiple projects simultaneously.
The existing system couldn't scale. Query times degraded as event volume grew, dashboards lagged behind real-time, and there was no data layer capable of feeding the AI analytics models the team wanted to build for predictive maintenance and anomaly detection.
Key challenges
- Millions of IoT events per day from hundreds of assets
- Sub-second query requirements for live dashboards
- Existing monolith couldn't handle the event volume
- No data architecture suitable for feeding AI models
- 99.9% uptime required. Construction can't pause.
- Multi-project, multi-tenant data isolation required
Our approach
- Microservices Rails architecture, each domain independently scalable
- Elasticsearch as the real-time event query engine
- PostgreSQL for transactional asset and project data
- AI-ready data schema designed for downstream ML model consumption
- AWS EC2 + RDS + S3 with auto-scaling and CloudWatch monitoring
- Docker containerisation for consistent, reproducible deployments
The AI Analytics Layer
The backend wasn't just built to serve dashboards. It was architected specifically to feed AI models. The data schema, event taxonomy, and Elasticsearch index design were all planned with downstream ML consumption in mind from day one.
Elasticsearch indexes are structured so that predictive maintenance models can query asset behaviour patterns, detect anomalies in real time, and surface alerts before failures occur, all running continuously against live event streams without impacting dashboard query performance.
🔮
Predictive Maintenance
AI models analyse usage patterns and sensor data to predict equipment failures before they happen, reducing unplanned downtime on construction sites.
⚠️
Anomaly Detection
Real-time anomaly detection flags unusual asset behaviour such as engine overheating, unexpected location changes or usage outside operating parameters.
📊
Utilisation Intelligence
AI-powered utilisation analysis surfaces which assets are underused, over-deployed or showing degradation, optimising fleet allocation across projects.
System Architecture
Real-time IoT event pipeline
🚛
Fleet Assets
IoT sensors
→
📡
Event Ingestion
Rails microservice
→
🔍
Elasticsearch
Real-time index
→
🤖
AI Models
Predict · detect
→
📊
Live Dashboard
Site managers
Infrastructure layers
1
IoT Ingestion Layer
Ruby on RailsMicroservicesEvent streamingMulti-tenant
2
Search & Query Engine
ElasticsearchSub-second queriesAI-ready indexesReal-time
3
Transactional Database
PostgreSQL (RDS)Asset schemaProject isolationAudit logging
4
AI Analytics Layer
Predictive MLAnomaly detectionUtilisation modelsLive inference
5
Cloud Infrastructure
AWS EC2AWS RDSDockerAuto-scalingCloudWatch99.9% SLA
Ruby on Rails
Microservices
Elasticsearch
PostgreSQL
AWS EC2
AWS RDS
AWS S3
Docker
CloudWatch
REST APIs