Ailee

Vision Intelligence for Non-Woven Bag Quality Control

year

2026

industry

INDUSTRIAL INTELLIGENCE

Client

Araliya Packaging Lanka (Pvt) Ltd

Secure air-gapped AI system for critical infrastructure

Overview

Araliya Packaging Lanka needed to inspect over 150 bag designs for more than 100 known defect types, at production speeds no manual team could sustain. We built Ailee, an AI-powered defect detection system that inspects up to 4 bags per second using dual industrial cameras and a two-stage deep learning pipeline, sorting defective bags in real time without slowing the line.Ailee brings high-speed optical scanning, automatic defect sorting, flexible pattern recognition, and real-time operator dashboards into one unified system helping packaging manufacturers eliminate manual inspections and protect product standards.

CHALLENGES

Manual inspection couldn't scale with production.

With production lines running past 1 bag per second, Araliya's manual quality process was already stretched thin, and it wasn't going to get easier as the catalog kept growing.

Slow Inspection Rates

Manual inspection took 10–15 seconds per bag, creating a hard bottleneck well below the line's actual output.

Labor-Intensive Process

A dedicated team of 15 staff was needed around the clock, driving up cost and exposing the process to fatigue-based errors.

Design & Defect Variability

Over 150 bag designs and 100+ defect categories, from print misalignment to fabric shading, made consistent manual inspection difficult to sustain.



150+

150+

Unique bag designs in production

100+

Known defect categories

15

Staff required for round-the-clock manual inspection

1015 sec

Manual inspection time per bag, before Ailee

The Approach

Match the hardware to the line speed, and let the software adapt.

Ailee started from a hardware constraint: to inspect at the line's real speed, the system needed to capture both sides of four bags at once, 8 images per cycle, using synchronized high-speed cameras, custom lighting, and precision mechanical feeding.

The software then had to keep pace without retraining every time a design changed. A two-stage pipeline separates general anomaly detection from targeted defect classification, so new SKUs can be onboarded without retraining the base model — a hybrid approach built to scale with Araliya's catalog, not fight it.



Match the hardware to the line speed, and let the software adapt.

Ailee started from a hardware constraint: to inspect at the line's real speed, the system needed to capture both sides of four bags at once, 8 images per cycle, using synchronized high-speed cameras, custom lighting, and precision mechanical feeding.

The software then had to keep pace without retraining every time a design changed. A two-stage pipeline separates general anomaly detection from targeted defect classification, so new SKUs can be onboarded without retraining the base model — a hybrid approach built to scale with Araliya's catalog, not fight it.



Custom enterprise software and AI automation dashboard
Custom enterprise software and AI automation dashboard
Custom enterprise software and AI automation dashboard

Core Capabilities

01  Dual-Camera Imaging & Precision Mechanics

Captures front and back images of 4 bags simultaneously (8 images per cycle) using custom LED lighting tuned for matte, non-woven surfaces.

02  Two-Stage AI Pipeline

Anomaly detection (PaDiM) flags deviations from a reference design; Vision Transformer classification then categorizes the specific defect type.

03  Real-Time Sorting & Actuation

Pneumatic diverters route defective bags off the line within milliseconds, with a fail-safe default to reject on any uncertain read.

04  Modular Admin Interface

Lets operators register new bag designs, review flagged defects, and monitor system health without engineering support.

Core Capabilities

01  Dual-Camera Imaging & Precision Mechanics

Captures front and back images of 4 bags simultaneously (8 images per cycle) using custom LED lighting tuned for matte, non-woven surfaces.

02  Two-Stage AI Pipeline

Anomaly detection (PaDiM) flags deviations from a reference design; Vision Transformer classification then categorizes the specific defect type.

03  Real-Time Sorting & Actuation

Pneumatic diverters route defective bags off the line within milliseconds, with a fail-safe default to reject on any uncertain read.

04  Modular Admin Interface

Lets operators register new bag designs, review flagged defects, and monitor system health without engineering support.

Core Capabilities

01  Dual-Camera Imaging & Precision Mechanics

Captures front and back images of 4 bags simultaneously (8 images per cycle) using custom LED lighting tuned for matte, non-woven surfaces.

02  Two-Stage AI Pipeline

Anomaly detection (PaDiM) flags deviations from a reference design; Vision Transformer classification then categorizes the specific defect type.

03  Real-Time Sorting & Actuation

Pneumatic diverters route defective bags off the line within milliseconds, with a fail-safe default to reject on any uncertain read.

04  Modular Admin Interface

Lets operators register new bag designs, review flagged defects, and monitor system health without engineering support.

THE RESULTS

14,400/hr

14,400/hr

Bags inspected per hour (4/sec sustained)

Bags inspected per hour (4/sec sustained)

>95%

>95%

Defect detection and classification accuracy

Defect detection and classification accuracy

80%

80%

Reduction in manual quality inspection labor

Reduction in manual quality inspection labor

More Projects

More Projects

AI-powered strategic planning system developed by Xleron

Ideascale

Turning Months of Business Planning Into Minutes, Without Losing Trust

IdeaScale Japan needed to compress a business planning process that traditionally took months, without handing sensitive corporate strategy to AI tools known to leak data or invent facts. We built IdeaScale Intelligence, an enterprise-grade AI platform that functions as a digital business advisor — running market research, competitor analysis, and financial planning inside a zero-retention, citation-enforced environment built specifically for confidential strategic work.

Secure air-gapped AI system for critical infrastructure

Sentrix

Securing the Heartbeat of Sobadhanavi's Power Generation

Sobadhanavi runs mission-critical national power infrastructure where a single missed diagnosis can cascade into a grid-wide outage. We built Sentrix, an air-gapped, on-premise AI architecture that turns live telemetry, field observations, and decades of engineering manuals into instant, actionable root cause analysis — without a single byte of plant data leaving the facility.

Conversational AI booking assistant for businesses

TabiAI

Redefining the Pre-Trip and In-Trip Experience for Hankyu Travel

Hankyu Travel needed its premium booking experience — including the Royal Collection — to match the level of care its human concierges provide. We built TabiAI, a conversational AI concierge that replaces static forms with natural dialogue, generates personalized itineraries in real time, and stays with travelers from first inspiration through in-trip navigation.

AI-powered strategic planning system developed by Xleron

Ideascale

Turning Months of Business Planning Into Minutes, Without Losing Trust

IdeaScale Japan needed to compress a business planning process that traditionally took months, without handing sensitive corporate strategy to AI tools known to leak data or invent facts. We built IdeaScale Intelligence, an enterprise-grade AI platform that functions as a digital business advisor — running market research, competitor analysis, and financial planning inside a zero-retention, citation-enforced environment built specifically for confidential strategic work.

Secure air-gapped AI system for critical infrastructure

Sentrix

Securing the Heartbeat of Sobadhanavi's Power Generation

Sobadhanavi runs mission-critical national power infrastructure where a single missed diagnosis can cascade into a grid-wide outage. We built Sentrix, an air-gapped, on-premise AI architecture that turns live telemetry, field observations, and decades of engineering manuals into instant, actionable root cause analysis — without a single byte of plant data leaving the facility.

Conversational AI booking assistant for businesses

TabiAI

Redefining the Pre-Trip and In-Trip Experience for Hankyu Travel

Hankyu Travel needed its premium booking experience — including the Royal Collection — to match the level of care its human concierges provide. We built TabiAI, a conversational AI concierge that replaces static forms with natural dialogue, generates personalized itineraries in real time, and stays with travelers from first inspiration through in-trip navigation.