Frequently asked questions
Short answers on document processing, workflows, and assistants.
What does Volerai do?
Volerai brings together classifying documents, extracting information, verifying results, and transferring them into business processes. It can be adapted to different document operations with organization-specific models and workflows.
Is it only OCR?
OCR is one of the steps used to read text in a document. Volerai combines that step with classification, field extraction, business rules, human review, and system integration.
Can we create a model for our own document types?
You can use your own document samples and labels for classification and extraction. The suitable model, training data, and success measures are set for the document type.
Does every agent use the same model?
No. Artificial intelligence models can be connected, and local models can be used within the installation. Each agent can use a different model. Which data goes to which model is decided during setup.
Is every result applied automatically?
The process moves according to the rules and permissions you define. Results that need checking can be routed to human review. How an assistant suggestion is applied also depends on the scenario.
Can it connect to our current systems?
Integration can be set up with API, database, ERP, and supported document-management connections. Verified data can also be transferred to other products you use. Access, data structure, and transaction rules of the target system are reviewed during installation.
Can it run inside our organization?
There are on-premises installation and local model options. Infrastructure needs, model files, and external connections are planned according to the chosen scope.
Is our data sent to external services?
That depends on the model provider and the connections in use. The installation decides which data is sent to which component, and the setup is reviewed against your data policy.
How are accuracy and speed determined?
Document type, scan quality, the selected model, and hardware affect the result. Field accuracy, processing time, and the need for human review are evaluated together on representative samples.
Can a model be trained on a few examples?
Yes. You label fields on the document and can start training from a small set of examples. How many examples are enough depends on the document type.
Can we decide who uses each agent?
Yes. Actions are written to a detailed log. Roles separate viewing from running, and you can define who may use each agent. Human approval can be part of the check step in the scenario.
Do we have to write code for a workflow?
The flow is designed without code. When one step needs more than the visual pieces, that step can be added in Java or JavaScript. It is an extra path beside the no-code flow.
How do we start?
Request a demo and share the document types you process, an approximate volume, and the target process. Scope is then set with suitable samples and acceptance measures.