The Company

Parascript LLC is a pioneer of cognitive capture software. Founded 25 years ago by data scientists and programmers who emigrated to the US from Russia, and based in Longmont, Colorado, Parascript earned its place as a leading handwriting recognition software for mission-critical business processes.

If you lived in the US during the past 25 years, you likely benefited from Parascript’s technology: the United States Postal Service (USPS) uses Parascript to read envelopes on its ultra-high-speed sorting machines for delivery to the right zip code; Parascript leveraged its technology to become a leader in check-reading software after the Check21 act was passed, helping to reduce the time for funds from a check deposit to appear in your bank account; and in the 2020 election cycle, Parascript signature verification validated millions of vote-by-mail ballots.

As handwriting is by far the most difficult text for computers to read, Parascript had to develop advanced algorithms and use machine learning (ML) from day one to train the computer to read better and faster. Building on the strength of its core AI and non-text-recognition technologies, Parascript branched out from handwriting recognition into general document classification and data extraction applications. Today, it offers a full complement of classification and recognition tools.

The company’s business model is to sell its suite of software tools through a global network of system integrators and value-added resellers, who integrate the tools into a variety of business processes. Recently, Parascript has focused more development on two document-intensive markets with legendary problems: mortgage and loan processors, and healthcare insurance companies.

The Technology

Smart Learning is the label Parascript uses to describe its approach to machine learning. To differentiate itself from mega AI platforms such as Google, AWS, and Microsoft, Parascript describes its product as an “artisanal” ML platform purpose-built for document processing.
Different document tasks require different combinations of algorithms, and the system is smart enough to choose what’s best. Most tasks involve multiple ML algorithms to overcome problems such as sample set size or bias potential. As a result, Parascript now has over 25 ML algorithms to throw at a document, including convolutional neural networks (CNN), recurrent neural networks (RNN), support vector machines (SVM), and several bespoke image-processing algorithms.

Parascript recently launched version 8 of its flagship product FormXtra.AI with a bevy of new cognitive capture features. Two in particular stood out to us from the demo.

1. Single pass page-level transcription of text and handwriting. This means Parascript’s software can read both machine text and handwriting in the same pass. This is a breakthrough that was not possible until recent innovations combined with increasingly inexpensive computing power. We saw an impressive demo of a handwritten note in a medical chart that also contained text. Single pass is at least 100% faster than the customary two passes. One potential use case is to process data for the Healthcare Effectiveness Data and Information Set (HEDIS), a comprehensive set of standardized performance measures designed to provide purchasers and consumers with the information they need to reliably compare health plan performance. This application typically includes medical charts, which are notoriously difficult for computers to read (see Figure 1).
   
2. Natural language processing (NLP) context to convert unstructured documents with non-tagged data into tagged data (see Figure 2). Parascript sees this as a game-changer for two industries:

— Mortgage and lending: for example, extracting the legal description from grant deeds that come in various formats. Parascript pre-trained the system for the example we saw.

— Healthcare: policies and correspondence, especially with handwriting.

Parascript also has interesting plans for the future. Harnessing the power of deep learning, the company is working on a new signature verification tool that can validate a signature on a loan document without needing a signature database for comparison. That is an impressive breakthrough. Also in the works is an algorithm that can detect and verify a notary stamp on a legal document, a notoriously difficult-to-read image found in every loan origination process.