PRR Enterprise Vision AI · July 2026
Starkey Vision AI · Enterprise Platform

From one inspection
to a vision platform across manufacturing.

Your leadership asked a bigger question: can this scale across manufacturing? This deck covers how the platform handles product variability, new products, added inspection points, governance, and the performance metrics your IT team wants.

Prepared for · Starkey Manufacturing leadership In response to · the VP inspection-strategy review Site · Matamoros, MX
What changed

What your leadership is really asking.

The POC proved the hardest part: telling apart subtle cosmetic differences under real line conditions. Every other inspection you raised runs on that same foundation. Your six topic areas are the right questions to settle before scaling. This slide maps where each one is answered.

TOPIC 1 · SLIDE 4

Color & variability

Shades, black tolerance, confidence, pink, and the colors that carry more risk.

TOPIC 2 · SLIDE 5

Shape & evolution

Revisions, new generations, and exactly what retraining each takes.

TOPIC 3 · SLIDE 6

Beyond cosmetic

Buffing, assembly, and additional checkpoints onboarded onto the platform.

TOPIC 4 · SLIDE 9

Future readiness

New-product onboarding, maintenance cadence, and per-product rules.

TOPIC 5 · SLIDE 10

Metrics

Confusion matrix, precision/recall, false-positive and false-negative rates.

TOPIC 6 · SLIDE 11

Enterprise vision

Your full inspection inventory, assessed point by point for feasibility and effort.

The shift: you're evaluating whether PRR can run the vision layer across your plant. That is what the platform is built for.
The reframe

The platform is the product. Each inspection plugs into it.

Each inspection is its own model: cosmetic today, buffing or assembly next. They all sit on one shared foundation for capture, edge inference, deployment, governance, and reporting. A new inspection inherits all of it on day one. That reuse is what separates a platform from a one-off build.

Color / cosmetic
proven · POC
Assembly verify
proven · POC
Serial OCR
proven · POC
Buffing
candidate
Post-spine assy
candidate
Missing parts
candidate
▲ each inspection is a model that plugs into ▼
The shared Vision AI platform, built once and reused for every inspection
Capture & lightingcameras · fixtures · edge box
Edge inferencelive PASS/FAIL on the line
MLOps & versioningtrain · deploy · roll back
Governanceapproval · drift · audit
Reportingmetrics · dashboards
Topic 1 · color recognition & product variability

How the model handles color variation.

The model learns the decision boundary between colors from labeled examples, captured across the variation your line actually produces. It does not rely on fixed RGB values, which is why it holds up where a rules-based system fails.

SUBTLE SHADES

Same family, different shade

We collect samples across the full range a shade actually shows on your line, so the model learns to separate adjacent tones instead of memorizing one "ideal" swatch.

BLACK-ON-BLACK

Controlled lighting plus real variation

We handle low-contrast black variation with controlled lighting (diffuse, grazing, cross-polarizer) and training on real lot, lighting, and material differences. Where two blacks read as identical to a calibrated sensor, we flag it and route to a person rather than invent a distinction.

CONFIDENCE

Every verdict carries a score

Each classification reports a per-class confidence. Below an agreed threshold the unit routes to a person for a re-scan. That behavior produced zero escapes in the POC.

PINK & RISK FAMILIES

We name the risk up front

Neutral tones (blacks, grays), closely adjacent shades, and glare-prone metallics carry more risk. Pink against rose or skin tones is handled the same way: train on the confusable set, then validate that separation. A color-risk assessment names these before deployment.

Validation approach: seeded edge-case test sets plus a confusion matrix on the risky color pairs, with accuracy thresholds agreed with your quality team before go-live.
Topic 2 · product shape & design evolution

Product changes, and the retraining each one needs.

Retraining effort scales with how much changed. In every case it is a data task, not an engineering rebuild, and the architecture stays the same.

MINOR REVISION

Add samples, fine-tune

A small design tweak or a new finish: add labeled examples, fine-tune, re-validate. The existing model absorbs it in days.

NEW GENERATION

Collect & retrain that product

A significant redesign gets its own image set and a retrain, or a new product class. The same pipeline we proved in the POC, pointed at the new geometry.

IMAGE VOLUME

Predictable, defined up front

Typically hundreds to low-thousands of labeled images per new variant, depending on complexity. We set the exact number per product so onboarding effort is never a surprise.

The New Product Introduction (NPI) playbook makes this repeatable: a defined image count, a fixed collection and training window, a validated model. When engineering ships a new product, the inspection follows a known path.
Topic 3 · expansion beyond cosmetic inspection

Adding an inspection reuses the platform underneath.

Each candidate you raised is a new model on the same platform. We assess every one through the same four-question lens, so "is this feasible?" gets an evidence-based answer.

CANDIDATE

Buffing verification

Surface finish and texture consistency in Custom Manufacturing. Feasible as a surface-anomaly model; it likely needs raking light and higher resolution, which we scope in the assessment.

CANDIDATE

Post-spine assembly

Presence, correct component, orientation. This is the closest to proven work: assembly-mismatch detection was one of the three checks the POC validated.

CANDIDATE

Missing components & more

Missing-part detection, additional QA checkpoints, OCR at new points. Each is a model; the platform underneath is already built.

The feasibility lens, applied to every inspection: ① Is it visually detectable? ② What optics and lighting does it need? ③ What data must be collected? ④ What is the effort to build and validate? Those four answers, per inspection, are the core deliverable of the roadmap engagement (slide 12).
The operating loop

Every inspection follows the same lifecycle.

One repeatable path from a new inspection to a governed production model, looping back whenever your business changes.

01

Collect

Representative images from the line.

02

Label & train

To your spec, on your data.

03

Validate

Held-out test, agreed thresholds.

04

Deploy

Live at the edge, human in loop.

05

Monitor

Drift & accuracy, continuously.

06

Retrain

When drift or change appears.

07

Version

Registered, approved, reversible.

NEW COLOR

Re-enter at Collect: add the shade, fine-tune, re-validate. The rest of the loop is unchanged.

NEW GENERATION

Re-enter at Collect with the NPI image set, retrain that product, validate, deploy.

NEW INSPECTION

A fresh model runs the loop once, then joins the platform's monitoring and governance.

Scalability

Each inspection you add costs less than the last.

The first inspection pays to build the platform. Every one after it reuses the capture rig, the edge deployment, the MLOps, the governance, and the dashboards. What is left to build is the model and its data.

LIVE

Cosmetic

Builds the platform.

NEXT

Assembly

Reuses everything.

NEXT

Buffing

+ model & data.

NEXT

OCR / missing parts

+ model & data.

Enterprise-wide

One platform, many checks.

Once
You build the platform once: capture, deployment, MLOps, governance, reporting
Per add
Onboarding a new inspection = a model + its data, on the foundation that already exists
Topic 4 · governance & long-term ownership

A production operating model you can own.

VERSIONING

Every model is registered

Each version is tagged with its training-data hash, its validation metrics, and its approver. You can see exactly what's running and roll back to any prior version instantly.

RETRAIN CADENCE

Driven by drift, not a calendar

Continuous monitoring flags when accuracy drifts or a new variation appears. That flag is the retrain trigger. You retrain on evidence, not on a fixed schedule.

APPROVAL & HUMAN-IN-LOOP

Nothing goes live unapproved

A new or retrained model passes validation and a human sign-off before it ever reaches the line. Your quality team stays release-of-record; the AI advises.

PER-PRODUCT RULES

Independent logic per product family

Inspection criteria and thresholds are configurable by product family, so different products carry different rules and models. Rule management scales through platform config instead of new code.

Who does what: PRR builds, retrains, and governs the platform and its models. Starkey owns the data and the release approval, and runs day-to-day operations after handover, with PRR on-call for retrains, new products, and upgrades. Drift detection, rollback, and a full audit log are built in from day one.
Topic 5 · inspection performance & quality metrics

The confusion matrix your IT team asked for.

We measure every model on a held-out set with the full confusion matrix. Those four numbers produce every metric your data team wants, and we set the operating point with you.

Predicted
DEFECT
Predicted
OK
Actual
DEFECT
TPcaughtFNescape, costly
Actual
OK
FPfalse alarm, re-scanTNpassed
Precision
TP / (TP + FP)
of flagged, how many were real
Recall / detection rate
TP / (TP + FN)
of defects, how many we caught
False-positive rate
FP / (FP + TN)
false alarms → a re-scan, costs seconds
False-negative rate
FN / (FN + TP)
escapes, the costly miss we minimize
The POC already ran at this operating point: ~6% flagged for a precautionary re-scan (false positives we accept) and zero escapes across ~2,300 units (false negatives driven to zero). Moving the confidence threshold trades those two against each other. We set that point with your quality team, then monitor it live with sampled human validation and drift dashboards.
Topic 6 · enterprise-wide inspection vision

You map the inspections. We map the path for each one.

Daniel and Citlali are building the beginning-to-end inspection inventory. For every point on it, we return a structured assessment, so leadership gets a costed blueprint to decide against.

YOU PROVIDE
  • 1The enterprise inspection inventory, every checkpoint front to back.
  • 2Representative images from each candidate point.
  • 3Priority ranked by business impact.
WE RETURN, PER INSPECTION
  • Can the platform do it? (feasibility, with evidence)
  • What the solution architecture looks like.
  • Level of effort, data collection, and training required.
  • How it phases into the rollout.
That assessment across your whole inventory is the roadmap engagement on the next slide. It turns an inspection list into a costed, sequenced plan.
How we get there

The path from here to enterprise scale.

STEP 1 · now

Enterprise strategy alignment

This deck and the next working session, to align on the platform vision, the architecture, and the path to production. No cost.

STEP 2 · 4–6 wks

Production Planning & Enterprise Vision AI Roadmap paid engagement

A fixed-scope engagement that produces your blueprint: production architecture · product-onboarding & retraining strategy · model governance & maintenance · performance-metrics & reporting framework · a prioritized inspection roadmap across your inventory · a phased production SOW. It gives leadership a costed plan before any large commitment.

STEP 3 · phased

Phased production rollout

Phase 1: productionize cosmetic inspection, live on the floor. Phase 2: expand to additional inspection points (assembly, buffing, OCR, missing components). Phase 3: scale the platform across manufacturing with ongoing support, governance, and enhancements.

Why this order: the roadmap gives leadership a costed, sequenced plan before any large production commitment, so every dollar after it is spent against a blueprint you have already approved.
Next steps

What each side brings to the next session.

FROM STARKEY

The inputs

  • 1The enterprise inspection inventory (front-to-back checkpoints).
  • 2Representative images from the priority inspection points.
  • 3Priority ranking by business impact.
FROM PRR

The guidance

  • Scalability & product-onboarding strategy.
  • Model maintenance, governance & retraining approach.
  • Feasibility read on your candidate checkpoints.
  • The performance-metrics & reporting framework, and the roadmap-engagement SOW.
Send the inspection inventory, images, and priority ranking. We'll bring the feasibility read and a roadmap SOW to the next session.— PRR · next step
01 / 13
← → or ↑ ↓ to navigate