Everything your team will ask.
In one place.

VeroLab is a tabletop device and software platform for analyzing visible physical characteristics of green coffee, including defects, screen-size distribution, and color, and for estimating bean volume and density. Results are presented in a web application, where you can inspect the individual beans behind the analysis.

The device measures 38 × 34 × 24 cm, approximately the size of a small printer, and weighs 15 kg.

You can register a sample on the device touchscreen or through the web application on a computer or mobile phone. Enter the sample ID, origin, processing method, sample weight, and other relevant metadata. With an established API integration, some or all of this information can be retrieved from your ERP system.

Yes. A scale and barcode scanner are included in the standard package.

VeroLab can analyze samples ranging from a few beans to 500 g. You enter the weight separately for each sample. The recommended sample size is 300–350 g, consistent with most quality standards.

A typical 300 g sample takes approximately 4–4.5 minutes. The exact time depends on the number and size of the beans, as well as the amount of small particles and foreign material, because each bean must be separated and captured individually. Low-grade samples with smaller beans and more particles may take 7–8 minutes.

There is no fixed limit on the number of samples that can be analyzed consecutively in one day. Practical throughput depends on the sample size, analysis time, loading and registration time, and your team’s working schedule.

No. VeroLab analyzes and documents the sample, creating a traceable record for each bean. It does not physically sort or separate the beans; they all exit together.

After each sample, VeroLab runs an automated cleaning cycle. A vacuum system cleans the imaging area and removes dust and residue. Periodic manual cleaning is still required.

As a general guideline, manual cleaning is required once every two to three months, depending on the volume and dustiness of the samples analyzed. A trained member of your team can perform the cleaning by following Agrivero’s instructions.

The AI model evaluates the available views of each bean separately and generates confidence scores for possible defect categories. The aggregation layer then combines the view-level predictions and applies defined rules to produce the final bean-level defect label.

A confidence score indicates how strongly a view resembles a particular defect category based on examples in the training data. It does not indicate that the complete sample is “X% accurate.”

The AI model can identify several candidate labels for a single bean. The applicable aggregation and grading-profile rules determine the final label.

For example, a bean may resemble both a broken bean and a part-black bean. The relevant precedence or severity rules determine which defect label is ultimately applied.

• AI model: Identifies visible characteristics and possible defect classes independently of the selected grading protocol.
• Grading protocol: Determines how recognized defects are named, counted, and converted into Full Bean Equivalent.
• Grading profile: Applies agreed thresholds, precedence or severity rules, and internal specifications for a particular origin, processing method, product, or grading methodology.

Agrivero’s AI model is trained on beans from multiple origins, processing methods, and harvests. It is not limited to a single region or grading profile. Over several years, we have worked with qualified graders to annotate images and expand the training data.

You can use the feedback function available for each bean and select a more appropriate defect category.

No. Feedback is reviewed and incorporated into a later AI model update or grading-profile adjustment. Correcting a single bean does not instantly retrain the current model.

Yes. During onboarding, we can align thresholds and aggregation rules with your grading methodology. A self-service grading-profile editor is planned for a future release.

We distinguish between two different processes:

• Device calibration: Technical calibration of the physical device or camera system. We can manage this remotely.
• Grading alignment: Ensuring that VeroLab’s output matches your grading protocol, terminology, and internal methodology. This forms part of the onboarding workshop and the alignment process between your quality team and ours.

We verify VeroLab results by comparing them with manual grading across a wide range of samples using supported protocols, including SCA, New York, and COB.

We provide a defined validation and grading-alignment report using representative samples. A general guarantee that every result will match every human grader is not realistic because graders, terminology, and protocols may differ.

No. VeroLab can reduce repetitive physical grading work, but it does not replace cupping, professional judgment, or final commercial responsibility.

The core product includes:

• The VeroLab device.
• The AI Grading License, including: sample registration, standard sample results, bean-level review, supported reports and exports, and standard software and AI model updates.

We are developing optional software modules that use green coffee grading data in other areas of coffee operations:

• Cupping.
• Blending.
• Trading.
• Roasting.

These modules are not required to use VeroLab’s core AI grading functionality.

No. The Cupping Module provides a place for your quality team to record sensory results within the same sample record used for green coffee grading. Human cupping remains necessary.

Combining the data creates a single sample history and may help your quality team identify relationships or irregularities between physical and sensory results, supplier changes, and crop-to-crop variation.

Agrivero currently supports CQI- and CVA-style forms. Multi-user and custom-form functionality is still in development.

The Blending Module uses available lot data to visualize how different coffees could be combined to meet targets for screen sizes, defects, sensory profile, and financial outcome.

No. Your team remains in control. The software may provide visualizations or suggestions, but it does not make autonomous commercial decisions.

These are early-stage or roadmap modules that would use green coffee grading data for roasting integration and quality-to-price benchmarking. They are not part of the current core product.

VeroLab is shipped in a protective case. To install it, place it on a suitable workspace, connect it to power, and connect it to a stable Wi-Fi network.

We provide remote support for setup. On-site installation or training is optional but recommended.

Yes. We provide a power cable appropriate for the installation location.

We offer a modular two-day workshop covering:

• Review of your use cases for VeroLab.
• Introduction to AI technology and the VeroLab product.
• Portfolio sample testing and initial grading alignment.
• User training and workflow integration with the quality team.
• Requirements review and planning for future adjustments.
• Definition of customer-success metrics and KPIs.

Because the workshop is modular, different team members can join the sessions most relevant to their roles. We recommend participation from:

• The Head of Quality and members of the quality team.
• Buyers and traders.
• IT representatives.
• The Head of Sustainability, where applicable.

We first provide remote support and guide your team through troubleshooting. If the issue cannot be resolved remotely, our team will service the device in Berlin.

No. VeroLab is designed to support long-term software and AI model updates on the existing hardware.