Document Swarm Learning®: Autonomous AI technology leveraging mass-training data.
Parashifts key innovation is a set of features called Document Swarm Learning® AI. It is a technology that systematically collects data, retrains machine learning models and deploys improved models to the platform automatically. Learn more what Document Swarm Learning® is all about.
Parallelized learning using different methodologies leads to superior extraction quality.
Learning on field level
Parashift’s document AI trains on field-level entities and can scale training across documents.
Leveraging the swarm
Swarm predictions from a network of AIs drastically improve the outcome of prediction problems.
Independent AI agents
Independent AI agents specialized in a particular task achieve powerful performance.
A radically new approach to Intelligent Document Processing.
Parashift’s vision has always been to build a generic Intelligent Document Processin API that allows document-based processes to be automated in existing applications with as little configuration, training and integration as possible. Document Swarm Learning is the answer to the question of how to pre-train 3,000 document types in the future.
“We are building a lot of new technology to keep you, the customer, at the forefront of developments in intelligent document processing.”
Andre Bieler (Head AI Parashift)
Parashift platform collects training data from across all use-cases and tenants, converts this data into its proprietary training data format and leverages the large combined data-network to train thousands of models automatically.
A first-of-its-kind approach to solve document data extraction.
Contrary to conventional systems, Parashift Platform does not link learning models to specific document types. Instead, the platform splits models into extraction entities. Extraction entities represent a structured, normalized value to be found. In a software application, this can be a data field, for example.
Learning models decoupled from document types.
Parashift’s breakthrough methodology is applied independently of the extraction approach itself. The extraction entities are separated from the document types and decoupled from the learning models. This allows the extraction methods to be related to each other in a new way.
Multiple models for the same field entity.
Per extraction entity, learning models and extraction mechanisms can be placed in competition with each other. Another model maps the evaluation of the best learning model and extraction method in each case. In this way, it’s possible to improve learning models independently of the document type. The improvement leads to the fact that for the same extraction entities no additional learning effort has to be driven for the respective tenant. This procedure also enables the simple and time-reduced configuration of new document types with “pre-learned” extraction entities.