AI visual inspection · pharmaceutical manufacturing
Inspect every unit. Explain every deviation.
Phamqyra learns each format from good units, catches the particles, cracks and closure defects nobody labelled, and traces every deviation to the batch or component lot behind it — with a record QA can sign.
- Awaiting units…
No correlated pattern in this batch yet.
- 01Few-shot teaching
- A few dozen conforming units teach a new format. No labelled defect library, no rules recipe.
- 02Zero-shot deviations
- Particles, cracks, fill and closure deviations it was never shown — flagged on first occurrence.
- 03Batch-level root cause
- Every deviation tied to batch, line, station and component lot, with a signed audit trail.
Formats inspected
01 / Method
Learn. Detect. Investigate.
Three steps that replace a recipe engineer, a labelled defect library and a tired inspection booth on the night shift.
Teach a format from conforming units.
Capture a few dozen good units on your own station, with your own lighting. Phamqyra builds a self-supervised model of what a conforming unit looks like for that presentation — then locks it as a versioned recipe ready for qualification.
- Few-shot: conforming units only, no defect library
- One locked model per product, presentation and station
- Re-teaching is a change-controlled new version
02 / Investigation
From “reject” to “here’s why.”
Rejecting a unit is the minimum. Phamqyra charts every deviation class against batch and component events, and tells QA which change explains the shift — while the batch is still on the line.
03 / Comparison
Manual inspection tires. Rule-based AVI ossifies.
Phamqyra learns each product's normal and keeps learning under change control.
| Criterion | Manual visual inspection | Rule-based AVI | Phamqyra |
|---|---|---|---|
| Set-up for a new product | Qualify inspectors with defect kits | A recipe engineer per presentation | A few dozen conforming units |
| Deviation types nobody defined | Sometimes — attention dependent | Only what the recipe encodes | Zero-shot deviation detection |
| Consistency across a shift | Fatigue limits booth sessions | Consistent, until the product changes | Same model, every unit, every shift |
| False rejects of good product | Conservative by necessity | Tight thresholds reject variation | Learns conforming variation |
| Root cause for investigations | Inspector notes, after the fact | Pass/reject counts only | Correlated to batch and component lot |
| Records for batch release | Paper forms and signatures | Machine logs, limited context | Signed e-records with frame hashes |
04 / Compliance
Built for GMP, not retrofitted for it.
AI that QA can actually release: locked, versioned models, a validation package per version, and records that stand up to an inspection.
05 / Solutions
From fill-finish to the final carton.
- 01
Sterile injectables
Vials and ampoules, inspected container by container.
- Visible particulate and fibres
- Glass cracks, chips and scratches
- Fill level and meniscus deviation
- Stopper position and crimp/cap defects
- 02
Prefilled syringes & cartridges
Barrel, plunger and needle shield in one pass.
- Plunger position and tilt
- Air bubbles and particulate in solution
- Needle shield presence and seating
- Barrel scratches, cracks and inclusions
- 03
Solid dose & blister
Every cavity, every seal, at line speed.
- Missing, broken or double tablets
- Foreign tablets and colour mix-ups
- Blister seal voids and foil damage
- Capsule dents, splits and cap mismatch
- 04
Labelling & serialisation
Every lot, expiry and code read, verified and logged.
- Lot and expiry print legibility (OCR/OCV)
- 2D DataMatrix grade and content
- Label skew, flagging and wrinkle
- Wrong or missing leaflet
- 05
CDMOs & multi-product sites
Dozens of products per line, one validated method.
- Any of the above, per product
- Product-specific acceptance criteria
- Sponsor-specific deviation categories
- Changeover verification
06 / Pricing
From $200 per camera per month.
Priced as software, not as a capital project. False-reject recovery and fewer investigations show up within weeks — which is why pilots convert.
Per camera
Up to $800 depending on resolution, frame rate and station count.
- Few-shot teaching, unlimited products
- Zero-shot deviation detection and maps
- Inspection records with audit trail
- Reject signal to the machine over OPC-UA
- Edge appliance included during pilot
Per line
Most pilotsAll stations on a line, batch investigation and system integration.
- Everything in Per camera
- Batch and component-lot correlation
- Draft deviations to QMS, results to MES and LIMS
- Validation package per model version
- Named validation engineer during pilot
Site & enterprise
Air-gapped deployment, CSV support and SLAs.
- Everything in Per line
- Air-gapped and segregated networks
- Computer system validation support
- Cross-site deviation intelligence
- Uptime and response SLAs
Pilot pricing and volume discounts available. Camera pricing scales with resolution and throughput; we'll quote after a 30-minute line review.
07 / Questions
What QA asks first.
Validation, data integrity and how it fits the line you already run.
Yes — if the model is treated like any other configured system. Each Phamqyra model is locked at release, versioned with the units it learned from, and shipped with protocol templates (IQ/OQ/PQ), a risk assessment and a traceability matrix. A released model does not change in production; improvements arrive as new versions through your change control.
Start a pilot
One station, one product, a few dozen good units.
Pilots run alongside your current inspection, so you see what Phamqyra catches — and what it stops rejecting — before anything changes in your batch record.