STATION-01 · STEADEYE GXP v1.0.0OPERATIONAL
CAPTURE_ALL · 6 CAM · VERDICT OK

Pharmaceutical vision system built for GxP

Steadeye is a camera-based counting and inspection station for pharmaceutical and other regulated production. Up to six cameras, AI detection models trained on your product, and a batch record that holds up in an audit. It runs on one Windows PC, fully offline.

LIVE / CAPTURE — OPERATOR SCREEN
BATCH_OPENED

What a pharma vision system has to prove. Not just what it sees.

Most machine vision systems are judged by one number: how often they are right. In pharmaceutical manufacturing that is only half the job. A count or inspection result that is not attributable, time-stamped and protected against change is just a claim, and an auditor will treat it as one.

Steadeye was designed from the record outwards. Every capture produces annotated evidence images, per-camera results and a verdict. They are written into a batch record that closes with an electronic signature and is chained into a SHA-256 audit trail. Seeing is the easy part. Data integrity is the product.

CHECK_LOGIC · COUNT | DAMAGE

Two inspections. One station.

Every capture runs one of two check logics, selectable per product run and recorded in the audit trail.

Counting
The detected quantity per camera is compared with the expected quantity and returned as OK, SHORT, OVER or NO_TRAY. Typical use: completeness checks on trays, kits and containers before release or packing.
Damage detection
Units the model classifies as damaged mark the frame DAMAGED. The damaged units go into the batch record, the PDF report and the result sent to the line controller.
Evidence for every verdict
Each frame is stored with its detection overlay, so a reviewer sees exactly what the station saw, not just what it decided.
EVIDENCE FRAME — VERDICT SHORT
MODEL_LOADED · sha256 recorded

Your product. Your model.

Pharmaceutical products differ in shape, color, coating and packaging, so a generic detector is rarely good enough. Steadeye includes an annotation tool and in-tool training: label your tablets, capsules, vials or components, including a damaged class, then train a detection model on the station and register it in a SHA-256-sealed model registry.

Models are standard ONNX files. The SHA-256 of the active model is written into the audit trail at load and into every batch record, so you can always show which model produced which result.

ANNOTATION — DATASET
PLC_TRIGGER · READY / BUSY / DONE

Part of the line. Not a lab tool beside it.

A vision system only saves work if the line can use its result without someone copying numbers by hand.

PLC integration
Native drivers for the two most common industrial PLC families. A trigger bit starts the capture; result code and counts are written back with a Ready/Busy/Done handshake.
REST API for MES and HMI
The same capture and batch functions are available over a documented REST API, served by the station itself.
One PC, no cloud
Detection, web interface, database and drivers run in one process on a Windows 10/11 PC, on the GPU when present and on the CPU otherwise. Production data never leaves the machine.
Up to six cameras
USB and IP cameras (RTSP or MJPEG over HTTP) can be mixed, each with its own label and expected count.
SIGNATURE_APPLIED · meaning: APPROVED

Validation is part of the delivery.

Steadeye is a GAMP 5 category 4 configured product. It ships with a computerized system validation starter kit: installation, operational and performance qualification protocols plus a requirements traceability matrix, which your QA team can execute as-is or adapt to your SOPs.

Electronic signatures follow 21 CFR Part 11 practice, closed batches are immutable, and the audit trail can be re-verified at any time.

HELP · F1

Questions quality teams ask.

What is a pharmaceutical vision system?

A pharmaceutical vision system uses industrial cameras and image analysis to check products and packaging during production, for example counting units, detecting damaged tablets or verifying that a tray is complete. Unlike general machine vision, it must also meet GxP data integrity requirements: results need to be attributable, time-stamped, protected against change and reviewable.

Which products can Steadeye inspect?

Any small, discrete unit a camera can see from above in a tray, container or fixture: tablets, capsules, vials, ampoules, closures or medical device components. Detection models are trained on your own product in the built-in annotation and training tool.

Is Steadeye a high-speed inline inspection machine?

No. Steadeye inspects at defined capture points: a PLC, an operator or another system triggers a capture, and all cameras record and evaluate it, in under a second on a GPU. It is designed for in-process and pre-release checks where the evidence matters, not for rejecting single units on a high-speed conveyor.

Does it need a cloud connection or internet access?

No. Everything, including licensing, runs offline on the station. Other PCs on your local network can use the interface over HTTPS.

How is Steadeye validated?

It is delivered with IQ, OQ and PQ protocols and a requirements traceability matrix following a risk-based GAMP 5 category 4 approach. Your validation lead owns acceptance; the kit gives them a documented starting point instead of a blank page.