01
Instruct
Hand over the work instruction, acceptance spec, or checklist—or write the criteria directly.
AI visual inspection for high-mix manufacturing
Visual inspection that works like a trained operator.
Netra understands your quality criteria up front, then applies them to every product with the speed of software and the nuance of human judgment.
Software that runs on-site with your existing cameras. Start on historical images, in shadow mode.
Camera · top viewSKU PCU-48
PCU-48 / REV C
connector seating
label match
Inspection resultReview
Overall confidence84%
Label matches jobPCU-48 / REV C
PassRequired modules present3 of 3 identified
Pass!
Connector fully seatedRight edge appears slightly raised
ReviewThe connector is present and correctly oriented, but the visible gap is close to the acceptable example boundary.
Acceptance criteria
Connector must sit flush. A visible edge gap is not acceptable.
Subjective criterion understood
Plan compiled8 checks · 2 views
Applies your quality standardsTrained like a new inspectorExplains every judgmentAsks when it's unsureRuns on your site
Beyond detection
Detecting a condition is not the same as judging its acceptability.
A camera can tell you a scratch is present. A trained operator tells you whether it's acceptable—by applying the standard, weighing severity, and knowing when to ask. Netra works the second way.
Detection asks“Is a scratch present?”
Inspection asks“Is this scratch acceptable for Grade A?”
Netra is powered by vision-language models that understand your acceptance criteria during setup and apply the approved standard to each product. That is what lets it judge rather than just detect. It sharpens that judgment with OCR, detection, measurement, multiple views, and production context, then returns a structured, explained decision.
Quality standardRefurbished display · Grade A
WI-204 v7Minor housing scuffs are acceptable. No deep scratch may cross the active display or front bezel. Damage visible at arm's length requires Grade B.
Display surface intact
Housing scuff outside display
!Scratch crosses front bezel
Netra judgmentGrade B
The scratch location conflicts with the Grade A criterion even though the display itself is intact.
The operator standard
Six habits of a trained inspector, built into the product.
Netra replicates human judgment. It breaks down your complex visual standards into repeatable, measurable checks that you can test during a pilot.
A trained operator…
Netra…
Reads the work instruction before looking at the part
Starts from your criteria document—not a training dataset
Judges severity and context, not just presence
Answers “acceptable for Grade A?”, not “scratch: yes or no”
Handles a variant they've never seen, from the drawing
Evaluates new configurations against the same written standard
Says “I'm not sure” and calls a senior
Routes low-confidence cases to human review instead of guessing
Explains why a unit was rejected
Returns the reason and the evidence region with every judgment
Is corrected once and doesn't repeat the mistake
Turns every correction into a permanent validation case
Netra works alongside your inspectors, not instead of them. Think of it as a tireless junior inspector who has read every document and never skips a step—while your experienced people supervise it, review its uncertain calls, and stay the authority on the standard.
A different layer
More than a smart camera. More focused than a vision platform.
Netra does not need to replace every existing tool. It sits above cameras, models, and vision utilities as the layer that defines, validates, operates, and improves the inspection itself.
| Capability | Smart cameras | Vision AI platforms | Netra |
|---|---|---|---|
| Starts with | Camera job, rules, or examples | Dataset, model, and workflow | Criteria, instructions, and examples |
| Primary abstraction | Inspection tool or configured job | Model and processing pipeline | Versioned inspection plan |
| Strongest fit | Stable, fast, bounded checks | Custom vision applications | High-mix, contextual, subjective judgment |
| When criteria change | Retune or reconfigure the job | Update data, models, and workflow logic | Edit the plan and replay trusted cases |
| Typical output | Detection, measurement, pass/fail | Predictions and workflow outputs | Judgment, evidence, explanation, review state |
| Ambiguity | Threshold or reject logic | Application-specific handling | Native human-review workflow |
| Who works closest to it | Vision or automation engineer | Developer or ML engineer | Quality and manufacturing teams |
Smart cameras optimize execution of a known vision job.
Vision platforms help technical teams build vision applications.
Netra helps an organization teach, validate, and operate visual judgment.
Onboard, don't program
Bring Netra up to speed the way you'd train a new inspector.
Hand over the instruction, show examples, check its calls, put it on the station with supervision, and coach it when it's wrong. The difference: Netra keeps every lesson—and re-proves itself against past cases before any change goes live.
02
Teach
Show good, bad, and borderline examples. Mark the regions that matter and clarify subjective concepts.
03
Validate
Check its judgments against historical cases you already know the answer to—before it touches production.
04
Deploy
Put it on the station—cameras, folders, or APIs—starting in shadow mode alongside your current process.
05
Improve
Review its uncertain calls, correct them once, and keep each correction as future validation evidence.
Controlled intelligence
A disciplined inspector, not a loose agent.
Netra doesn't improvise on every frame. The approved inspection plan defines what it evaluates, what evidence it gathers, what it returns, and when a person must review the result.
Written criteria define acceptability, severity, and disposition.
Visual examples clarify concepts that words alone cannot capture.
Multiple views provide the angles required for a complete decision.
Vision tools support OCR, detection, segmentation, counting, and measurement.
Production context connects the image to variant, job, barcode, or MES data.
Confidence gating separates automation from cases that deserve a person.
Structured judgment
Disposition: ReviewReason · evidence · confidence · action
↑
Netra reasoning layerEvaluate evidence against the approved standard
↑
Vision
Criteria
Views
Tools
Context
Examples
Industries
Built for industries where products, standards, and visual conditions keep changing.
Netra thrives in complex environments. When your production involves high mix, shifting criteria, and subjective human judgment, traditional fixed inspection rules fail. Netra adapts.
Inspection beyond fixed AOI programs.
Adapt to product variants, box-build configurations, connectors, labels, accessories, and visible workmanship criteria without treating every change as a new model project.
Discuss this industryBuilt for configured-to-order products.
Evaluate assemblies against the actual job criteria, BOM, work instruction, and approved visual references—even when products are rarely built exactly the same way twice.
Discuss this industryConsistent grading for inconsistent inputs.
Reason about severity, location, completeness, repairability, and cosmetic grade across products that arrive in widely different conditions.
Discuss this industryAdapt across variants and subassemblies.
Support off-line and end-of-line verification, rework checks, routing and connector review, cosmetic standards, and low-volume changing lines.
Discuss this industryAlso relevant toAerospace & defense manufacturingMedical devices & regulated production
We start with non-safety-critical, human-assisted workflows and validate every deployment against your quality requirements.
Built for trust
Read about deployment and data handling Do not automate the decision until it can be explained and tested.
Quality teams need more than an impressive demo. Netra is built around review, traceability, regression evidence, and controlled deployment.
Human review is part of the design
Netra routes uncertain results to a person instead of forcing a confident-looking answer.
Every plan is versioned
Criteria, examples, output fields, and deployed versions remain explicit and traceable.
Changes face trusted cases
Re-run known good, bad, borderline, and historical escape cases before a revised plan is frozen.
Local-first deployment
The inspection stack runs on-site. Cloud models stay optional—off unless your data policy allows them.
Inspection runRUN-2026-06-1842
Human reviewed09:42:11
Plan v4 loaded · criteria hash recorded
09:42:12
Top and side frames captured
09:42:14
8 checks evaluated · 1 routed to review
09:44:02
Reviewer corrected connector result and added reason
09:44:03
Case pinned to trusted regression set
Where to begin
Choose the inspections where Netra's approach creates a real advantage.
Strong starting point
Netra is a good fit when…
- Products or acceptance criteria change frequently
- A person currently inspects against a document or examples
- The decision requires context, severity, or subjective judgment
- Structured outputs are needed beyond a simple detection
- A few seconds per decision is acceptable
- Shadow or assisted deployment is possible
×
Use a different approach first
Traditional vision may be better when…
- —Millisecond, continuous line-speed inspection
- —Micron-level metrology or highly specialized optics
- —A stable geometric rule already solves the task reliably
- —Immediate autonomous release of safety-critical hardware
A bounded first deployment
Bring Netra one inspection that still depends on human judgment.
Start with one product family, one criteria document, and a set of historical good, bad, and borderline images. Evaluate offline first, then move through shadow and assisted modes with measurable success criteria.
1
Historical evaluationRun the current standard against known images.
2
Plan refinementClarify criteria, examples, outputs, and review rules.
3
Shadow modeCompare Netra with the current process without controlling it.
4
Assisted useAutomate confident cases and route uncertainty to people.