Pilora · QA copilot
From a flagged frame to a likely cause
Pilora reads defect events from Veridex and writes them up in plain language, routes borderline calls, and looks across shift, batch and process data for what changed upstream.
Specification
- Inputs
- Veridex defect events and frames; MES and historian telemetry
- Models
- Vision-language and language models customized with NVIDIA NeMo
- Serving
- NIM and Triton with TensorRT-LLM; on-premises option on IGX (planned)
- Guardrails
- NeMo Guardrails; answers cite the events they use
What Pilora does
- 01
Plain-language defect reports
Type, severity, location and probable cause for each flagged frame or cluster.
Core - 02
Borderline calls routed
Auto-pass, auto-reject or human review, with every routing decision logged.
Core - 03
Looks upstream
Correlates defect clusters with SKU, shift, batch, lighting and process telemetry.
Core - 04
Corrective suggestions
Recommends what to check and alerts on trends before they become escapes.
Core - 05
Questions in plain words
Ask about defect trends across lines and plants and get answers grounded in your data.
Core - 06
Drift triggers retraining
Detects drift and proposes a Coreon retrain for approval (planned).
Planned
What Pilora produces
Scratch cluster on AX-204, Line A
- Window
- Shift B, 14:00–16:00
- Rejects
- 23 scratches
- Location
- Top face, near slot
- Correlates with
- Fixture 4 change at 13:40
- Confidence
- Medium
Suggested action. Inspect fixture 4 clamp pads for debris. Scratch position matches the clamp contact area.
Inside Pilora
Collect
Defect events and frames arrive from Veridex.
Ground
Process data and past reports are retrieved.
Reason
A vision-language model drafts the report.
Review
An engineer confirms or corrects it.
Specification
Items marked planned are in design with pilot partners.
| Item | Detail |
|---|---|
| Inputs | Veridex defect events and frames; MES and historian telemetry |
| Models | Vision-language and language models customized with NVIDIA NeMo |
| Serving | NIM and Triton with TensorRT-LLM; on-premises option on IGX (planned) |
| Guardrails | NeMo Guardrails; answers cite the events they use |
| Outputs | Defect reports, triage decisions, root-cause suggestions, alerts |
| Status | Agentic root-cause analysis is in design with pilot partners (planned) |
Pilora in the Viseena loop
- P01 · GenerateSynvex
Renders labeled defect images from CAD, materials and a defect taxonomy.
- P02 · TrainCoreon
Fine-tunes a shared defect model into a per-SKU adapter, closing the sim-to-real gap.
- P03 · InspectVeridex
Runs the model on edge hardware at the station, at line speed, on every part.
- P04 · ExplainPilora
Turns flagged frames into defect reports, triage and probable root cause.
- P05 · CommissionTwinix
Simulates cameras and lighting in a line twin before anything is mounted.
Uncertain and confirmed real frames from Veridex flow back to Synvex and Coreon. Pilora triggers retraining when it sees drift, and Twinix sends new scenarios to Synvex before a changeover.
About Pilora
Does Pilora make decisions on its own?
Only the ones you allow. Triage rules are set by your engineers, and root-cause suggestions always need a person to act.
Will it make things up?
Answers are grounded in your inspection and process data and cite the events they rely on. If the data does not support an answer, Pilora says so.
Which process data can it use?
Anything you connect: MES, historian tags, maintenance logs. Pilots usually start with shift and batch data.
Try Pilora on one SKU
Pilots start with one part, one defect list and accuracy measured on real frames.