Cold Storage & Refrigeration

Predictive Analytics from Low-Interference Cold-Storage Data

A containerized processing pipeline designed to analyze weekend “no-disturbance” sensor data and generate high-confidence anomaly and efficiency insights using a custom AI model.

Weekend
No-Disturbance Window
Clean Data
Uninterrupted Sensor Capture
Docker
Containerized Platform
Predictive
Anomalies & Efficiency Signals
The Challenge

Reliable Predictive Insights from Stable Operating-Window Data

A cold storage and refrigeration company needed predictive insights from sensor data — but only from periods where assets were truly undisturbed, so the AI model could learn stable baselines and detect abnormal behavior with higher confidence.

The challenge was to capture and process full sensor streams during the weekend “no-disturbance” window, then run a custom AI model to identify anomalies and efficiency patterns — without interfering with normal operations. Running the pipeline on weekends also reduced unnecessary weekday compute, but the primary goal was data integrity and signal quality.

Core Requirements

  • Lightweight Ruby-based application running as Docker containers in AWS ECS
  • Two core processes: dispatcher for scheduling and workers for processing
  • Trigger processing during the weekend window (Sat/Sun), aligned with low/no operational disturbance
  • Collect complete sensor streams and store both raw and processed outputs in InfluxDB
  • Custom AI model integration to detect anomalies and efficiency signals across assets

Processing Schedule

Aligned to weekend “no-disturbance” windows

Mon
Monday - Friday
Idle
Sat
Saturday 11:59 PM
Active
Sun
Sunday 11:59 PM
Active

Processing is scheduled to match low/no operational disturbance, producing cleaner inputs for AI analysis and more reliable anomaly detection.

Our Solution

Weekend “No-Disturbance” Analytics Pipeline

A minimal yet robust containerized architecture optimized for clean data capture and predictive analysis.

Dispatcher

Cron-driven orchestration

Runs during weekend windows
Creates one job per asset per run
Places jobs in SQS queue

Worker

Predictive processing engine

Pulls jobs from SQS queue
Invokes custom AI model
Stores results in InfluxDB

Technical Implementation

Custom Ruby/Rack Framework

Developed a lightweight, modular internal framework optimized for predictable batch execution and maintainable long-term evolution.

Containerized Services

Dispatcher and worker services packaged as Docker containers for AWS ECS deployment with controlled scaling.

InfluxDB Integration

Stores raw sensor streams and processed AI outputs to support dashboards, alerts, and historical comparisons.

Processing Workflow

1

Weekend Trigger

Dispatcher runs during the weekend “no-disturbance” window

2

Queue Asset Jobs

Creates one processing job per asset and queues it via SQS

3

Predictive AI Analysis

Worker invokes the custom AI model to detect anomalies and efficiency signals

4

Persist & Visualize

Stores outputs in InfluxDB for dashboards, alerts, and historical comparison

Technology Stack

Modern, Containerized Infrastructure

AWS ECS

Container orchestration for dispatcher and worker services

SQS Queues

Reliable job queuing and message passing between services

InfluxDB

Time-series storage for raw sensor data and processed analytics

Custom AI Model

Anomaly detection and efficiency insights from weekend baseline data

Development Stack

Ruby Rack Framework Docker AWS SDK InfluxDB Ruby

Infrastructure

AWS ECS Fargate SQS CloudWatch IAM Roles VPC Networking
Results & Impact

Cleaner Inputs, Stronger Predictions

Key Outcomes

Complete Weekend Sensor Capture

Collected full sensor streams during low/no-disturbance windows to establish stable baselines and reduce noise in the data.

Predictive AI Anomaly Detection

Applied a custom AI model to detect anomalies and highlight efficiency patterns across cold rooms and deep freezers.

Scalable Orchestration

SQS-based job orchestration capable of processing assets in parallel with controlled scaling during weekend runs.

Fast Analytics Turnaround

Processing outputs delivered quickly after the scheduled run, keeping dashboards and operational insights up to date.

Before & After Comparison

Before: Mixed Operational Data

  • Sensor data collected during active usage introduced noise
  • Harder to establish stable baselines for AI predictions
  • Continuous processing increased operational complexity

After: Weekend Off-Hours Batch Processing

  • Cleaner inputs from undisturbed assets
  • More reliable anomaly and efficiency signals
  • Batch execution concentrated into planned windows
  • Reduced unnecessary weekday compute as a side benefit

Need Reliable Predictive Analytics from Sensor Data?

We design containerized, event-driven pipelines that capture cleaner signals and turn them into actionable anomaly and efficiency insights.

Predictive Analytics

Batch, event-driven, anomaly detection

Cloud Infrastructure

AWS, ECS, Lambda, cost-aware scaling

Automation & Monitoring

Scheduling, workflows, observability

Explore More Success Stories

Rails Modernization

Phased Rails Upgrade

Upgraded Rails 5.2 → 7.2 with production-safe checkpoints and a phased rollout strategy.

Read Case Study
Hosting & E-Commerce

Unified Hosting Platform

Integrated multiple hosting systems into one platform, improving monetization and operations.

Read Case Study
Supply Chain & IoT

Real-Time Cold Storage Monitoring

Fault-tolerant IoT pipeline with custom protocol integration and live dashboards.

Read Case Study