Automating Quality Control at Scale

Development of an Enterprise IoT Edge Intelligence Platform for AI-Powered Visual Inspection

We built Exacto from the ground up, an enterprise IoT edge intelligence platform that replaces manual quality sampling with 100% automated visual inspection on live production lines. The platform connects edge devices, IP cameras, and PLCs to a centralized dashboard where ML models are deployed, inspections run in real time, and every result is tracked and audited.

Outcomes: device commissioning reduced from multi-day scripting to a single guided session, quality reporting shifted from weekly manual summaries to live dashboards, and compliance audit preparation cut dramatically with a built-in audit trail and on-demand PDF export.

The Challenging Part

Manufacturing environments offered no single standard to build against. Exacto had to unify heterogeneous hardware, different camera brands, GigE/USB/TCP-IP protocols, image formats, and PLC standards (OPC UA, Modbus), into one structured onboarding flow without losing device-level precision. Beyond hardware, inspection results needed to reach the dashboard the instant they occurred on the production line, requiring a dual-path architecture combining real-time polling and background cron-based sync.

• Heterogeneous edge hardware with no unified protocol or image standard across camera brands and PLCs
• Real-time inspection results required without cloud latency, ML inference running on-device, syncing centrally
• Millions of inspection records (cycle counters, batch IDs, model outputs, timestamps, images) outpacing relational query performance
• ML model lifecycle management, deploying the right model to the right device for the right part number without disrupting production
• Compliance requirements demanding complete, exportable audit trails for every user action across the platform
• Flexible data storage, some clients require AWS S3, others mandate fully on-premise NAS for data sovereignty

‍

Our Solutions

We delivered a full-stack IoT intelligence platform built around nine integrated modules. The device onboarding module guides operators through a 7-step wizard, from bare hardware registration to a fully operational ML inspection station, with live IP connectivity testing and resumable progress so no configuration is lost mid-session. The real-time production monitoring interface streams pass/fail results and live FPY calculations per camera every four seconds, with a lock-based override system that maintains quality gate control without stopping the line.

A centralized ML model library manages version tracking, file uploads, and device-level deployment mapping, eliminating the manual per-device file transfers that previously caused deployment errors. Apache Solr was integrated as a dedicated search layer to handle full-text and filtered queries across 1M+ inspection records. The inspection results and reporting module generates PDF exports ready for compliance reviews, while the built-in audit log captures every create, update, delete, and override action with user identity and timestamp, making audit preparation a one-click operation rather than a manual reconstruction effort.

‍

Results

• 100% inspection coverage across all production cycles, replacing manual spot sampling entirely
• Device commissioning reduced from multi-day manual scripting to a single guided session
• ML deployment errors eliminated, models previously copied manually per device are now centrally version-tracked and audited
• Quality reporting shifted to live dashboards, weekly manual summaries replaced by real-time FPY data filterable by shift, part number, and model
• Solr-powered search returns results across 1M+ inspection records in seconds, not minutes via spreadsheet
• Compliance audit preparation cut dramatically, complete traceable history available on demand with PDF export
• Silent device failures now surfaced immediately via heartbeat monitoring, previously discovered only when production data went missing
• Multi-camera configuration simplified, up to 6 cameras per device managed through one unified form, replacing per-camera manual scripts