FAQ
Answers to common questions about Searchdoc products and deployment.
Before deployment, answer quality can be tested using your own materials and questions from your day-to-day work. Answers are generated based on evidence in the materials. When evidence is insufficient, VELOAT distinguishes between what has been verified and what requires further verification.
You can upload PDF, DWG, RVT, and IFC files. Scanned PDFs are processed using optical character recognition (OCR), and recognition quality for tables and images depends on the quality and layout of the source material. Exporting analysis and review results to a file is not currently supported.
The total number of documents and storage capacity are determined by the installation specifications, deployment configuration, and contract terms, and we explain the specific limits during deployment consultation. ZIP (compressed) files are not currently supported. Instead, if you upload a project folder as is, its subfolder structure is preserved.
VELOAT is available as SaaS operated by Searchdoc or as an installation in your own environment. For customer-managed installations, we discuss the environment, such as standard on-premises, your own VPC, or restricted outbound connectivity through an allowlist or proxy. We determine the appropriate option based on who operates the service, external connectivity, security requirements, and support scope.
To search and analyze your materials and generate answers, the necessary text and drawing images, along with user questions and conversation content, are sent to external AI models. In the current configuration, a connection to external model services is required even when VELOAT is installed on your own servers.
If external connections are not allowed, we consider installing VELOAT and the AI models together inside your environment, and check model performance, server specifications, and the integration work required in advance.
VELOAT uses Claude-family LLMs and the Titan embedding model through AWS Bedrock, while the OCR model that reads text in documents and drawings is operated in-house. The embedding model converts your material into numeric vectors so that it can be searched. For installations in your own environment, access rights and usage quotas for the LLM and embedding model need to be prepared in advance, and the models and versions to be applied are confirmed together based on your deployment configuration.
A PoC validates whether VELOAT can be used for your actual review work with your own documents. We first define the work to validate, the types and volume of documents you will provide, and the environment the PoC will run in. We then agree on whether it is free of charge, its duration, and who covers installation and model usage costs. The overall deployment schedule depends on material preparation and your environment.