Innowave Tech
Start a Conversation
English
Menu

Explore Innowave Tech

Start a Conversation

iWave ZeroDefects / Vision AI

Exceed Rising Quality Demands with Vision AI that Adapts.

Your inspection goals guide the AI. Builder Agents bring together and tune the models needed to recognize defects, with training, testing and deployment managed in ZeroDefects.

Explore Builder Agents Request a Demo
The Platform in Motion
See the Ensemble Take Shape
Project-Specific EnsemblesFull MLOps + AutoMLAPI-Enabled Applications

Full MLOps + AutoML

One Project.
A Complete Lifecycle.

Every project brings its own data, ensemble and MLOps process together in ZeroDefects.

  1. 01

    Data Pipeline

    Prepare and process the project’s data.

  2. 02

    Ensemble Model

    Manage model composition and individual weights.

  3. 03

    MLOps Process

    Train, validate and test through the lifecycle.

  4. 04

    Data Repository

    Keep each project’s data in its own repository.

  5. 05

    Deployment

    Deploy the model for its intended application.

iWave Builder Agents / Scaling Expertise

The Expertise behind the Ensemble.

Designing the ensemble. Setting each model’s weight. This is where expertise and time concentrate. We equip Builder Agents with the domain skills and tools of data scientists, data engineers and manufacturing engineers to make this work scalable.

01 / Define the Outcome

Bring the Data. Describe the Outcome.

Create a project, upload sample images and explain what they show and what you need to recognize. That inspection goal guides the Builder Agents.

Sample Data + Manufacturing Context + Your Goal

The platform keeps datasets and their annotations with the project. Your description adds the manufacturing context the images cannot provide alone.

iWave ZeroDefectsBuilder Agent Demo
ZeroDefects workflow demonstration: sample images alongside a typed goal to find defects and reduce false rejects.
Describe the inspection outcome alongside the sample images.

02 / Design the Ensemble

Build the Ensemble around the Task.

Builder Agents apply data-science, data-engineering and manufacturing expertise to design the ensemble and its pipeline. The goal determines how the models work together.

Agent-Program-Agent / Expert Model Composition

Each project has its own ensemble, data pipeline and MLOps process. Model roles and relative contributions depend on the task.

iWave ZeroDefectsBuilder Agent Demo
ZeroDefects workflow demonstration: the Builder Agent creates anomaly, defect-detection and classification models connected to a weighted decision.
Watch the Builder Agent compose the models and connect the ensemble.

03 / Tune the Weights

Give Each Model the Right Influence.

Choosing the models is only part of the expertise. Builder Agents also set the individual pipeline-model weights around your data and desired outcome.

Domain Expertise / Composition + Individual Weighting

Designing the ensemble and setting its weights are the most expertise-intensive parts of the work. This is the customization the Builder Agents automate.

iWave ZeroDefectsBuilder Agent Demo
ZeroDefects workflow demonstration: individual model weights change while the combined weighting remains balanced.
Tune each model’s contribution around the inspection goal.

04 / Train & Validate

Turn the Design into a Tested Model.

Continue through training and validation within the project. Review the data, training configuration and session results before moving to further testing.

Full MLOps + AutoML / Reviewable Results

Model construction starts the lifecycle. Training, validation, testing and deployment remain connected in ZeroDefects.

iWave ZeroDefectsTrain & Validate
ZeroDefects training configuration and completed training session history.
Training configuration and session history stay with the project.

05 / Inspect the Result

See How the Ensemble Reaches Its Result.

Test the pipeline with inspection images and examine the output at each model. Use the evidence to evaluate the ensemble before deployment to its intended application.

Image → Model Outputs → Inspection Result

Use APIs to build the application around the qualified model, including inference, results storage, self-learning features, wafer maps and KLARF updates.

iWave ZeroDefectsInspect the Result
ZeroDefects hybrid pipeline test showing image outputs and the route through connected models.
Follow the image through the pipeline and inspect each model’s output.

06 / Adapt to a New Task

A New Inspection Goal. Expertise Applied Again.

Bring the next dataset and objective. Builder Agents revise the model composition and individual weights, then the changed ensemble goes through its own evaluation.

AI Scalability / Automating the Customization

The scalable element is the expert work of adapting the AI to each task. A new objective gets a new design and its own qualification.

Expertise Applied to the Next Task

New Data. New Objective.

Model Composition
Retain, replace or add models to suit the task.
Individual Weights
Rebalance each model’s contribution.
Qualification
Train, validate and test the changed ensemble.
Compare Two Inspection Goals
Agent-Program-Agent

Builder Agents design and tune the ensemble. You take it through training, validation, testing and deployment. The same expertise adapts it to the next inspection task.

AI Scalability / Expertise Applied Again

Change the Outcome.
See the Ensemble Change.

Builder Agents use the new data and objective to revise model composition and rebalance individual weights. The new ensemble gets its own evaluation.

A sailing crew changing course together, leaving a curved wake across the sea
Illustrative comparison

Model roles, count and relative weights are examples, not a prescribed architecture or measured result.

01 / Original objective

Recognize the classes you know.

Your project brief“Classify these wafer defects so our application can use the results in wafer maps and KLARF files.”
Sample data + context
Sample wafer images with reviewed defect classes.
Task-specific evaluation
Validate class decisions against held-out, labeled examples.
Intended output
Defect-class results for the application layer.
iWave Builder Agents

Composition and individual contributions are designed around classification.

Original ensemble / Candidate designIndividual weighting
  1. A

    Appearance model

    Selected

    Higher contribution
  2. B

    Detail model

    Selected

    Supporting contribution
  3. C

    Class-decision model

    Selected

    Higher contribution

Project-specific ensembleCandidate design → ready for training

Illustrative model placeholders and relative contributions.

02 / Changed objective

Find patterns that need a closer look.

Your project brief“Use these wafer images to flag unfamiliar defect patterns for review, including patterns outside our known classes.”
Sample data + context
Representative wafer images, operating context and reviewed outliers.
Task-specific evaluation
Test unfamiliar-pattern detection and false alarms on separate examples.
Intended output
Flagged inspection results for a review workflow.
iWave Builder Agents

Builder Agents revise the composition and rebalance contributions for the new objective.

Adapted ensemble / Candidate designIndividual weighting
  1. A

    Appearance model

    Retained · reweighted

    Lower contributionMarker: original contribution
  2. B

    Detail model

    Retained

    Supporting contribution
  3. D

    Unfamiliar-pattern model

    Replaces the class-decision model

    Higher contribution

A is reweighted. B is retained. C is replaced by D. The new task gets its own evaluation.

Project-specific ensembleCandidate design → ready for training

Illustrative model placeholders and relative contributions.

Builder Output → Your Model Lifecycle

A New Ensemble.
Its Own Qualification.

You take the adapted ensemble through training, validation and testing before deployment. APIs then connect it to your application’s inference, results and review workflow.

  1. Train
  2. Validate
  3. Test
  4. Deploy
Explore the project lifecycle

The Application Layer / Enabled by APIs

Your Process.
Your Application.

Build custom features on top of the platform: inference, results storage and self-learning capabilities. Connect the outputs to the work your operation needs to do.

For semiconductor applications, that can include updating KLARF files and wafer maps.

Explore the Wafer Inspection Story
Connect Inspection Results to Wafer Maps and KLARF

Your Expertise. New Possibilities.

Put AI to Work in Your Operation.

Request a Demo