Cullable.

THE FINAL WORD IN TEXT RECOGNITION

DESKTOP APPLICATION EARLY ACCESS PROGRAM

Cullable Cloud Power.

NOW ON WINDOWS AND MAC

People use text for lots of things. We’ll take a stab and assume all of those things should produce better than a 70% accuracy rate. Most OCR engines average about 70% accuracy rate and exist to embed produced text into a PDF or produce a text file and that’s the end of the story. We do that too, but we use Computer Vision instead of OCR and average well over 90% on medium to poor quality documents, even much higher in many cases. Still, for many, this leaves much to be desired.

The team at Cullable asks, what’s next? Downstream processes are important and Cullable’s adding new creative transformations and thoughtful, operational staples downstream from text recognition all the time. From language translation, DLP, automated redaction or summarization using convolutional neural networks like OpenAI’s GPT, all the way through production of .json files cataloging the x and y position of words and characters, or even the edges or “bounding boxes” of recognized words, helping developers, healthcare, knowledge management, legal and digital transformation initiatives in a number of industries. Cullable’s technologies power form automation and machine learning initiatives that harvest intelligence from legacy document formats, lower operational cost of healthcare or determine taxonomies and privilege access rules for enterprise data lake migrations. It’s just useful in so many ways besides ‘making this pdf searchable’.

We’re launching Cullable’s desktop app early access program with Text Recognition as the cornerstone of the solution. We plan on implementing all of Cullable’s downstream processes over time, with priority of feature releases being dependent on customer interest and intent. The text recognition module’s release is considered very stable and is packed with a variety of features, such as the ability to generate PDF files from images, document or page level text file (supports .OPT for boundary definition), or choose to have your text embedded in an output PDF file. It should serve as a solid foundation in which to build on. Cullable’s desktop app is intended to be licensed to an organization, with org-specific api keys being generated for unique users or departments throughout the company. Each app uses it’s respective API key. This allows Cullable to report on a department or user’s specific utilization and cost, or to roll up activity over time in order to chart and understand organizational usage, adoption and ROI. Cullable’s text recognition engine is offered at a low cost per page and volume discounts are always available.

On execution, Cullable bundles the source images or files and securely encrypts them in a container, transferring them to your organization’s private cloud storage bucket. From there, pre-processing operations are performed and images are shipped to Cullable’s API cloud. OCR is processed, records are re-assembled and automagically downloaded to your configured destination directory. We don’t see your files. We don’t retain any text or ‘operational state’. We only track usage and top level project naming conventions in order to report them back to you.

Capitalize on the raw power of Cullable’s cloud using a drop dead simple workflow that anyone can learn in minutes, producing the worlds fastest, most accurate OCR almost instantly (at a rate of up to 20,000 pages per minute).

A full list of Cullable’s current OCR and downstream features are included on the following pages. Where do you want us to go next?

NEXT GENERATION TEXT RECOGNITION AS A SERVICE.

The highest quality text recognition solution in the world. It’s Document AI powered by Computer Vision. There is no reasonable comparison available as a readily deployable solution on the market today. Speeds up to 10,000 pages per minute with extraordinary language support with a processing overhead heap of about 1GB of memory and surprisingly low CPU commit. Available through the web app, with a secured API and as a desktop app.

MACHINE TRANSLATION AS A SERVICE

High quality machine translation of created or provided text, as a service. Highly scalable. Can be easily pipelined from or to any other Cullable service. For example, Korean documents are OCR’ed, that output is pipelined into translation, providing the user a searchable pdf, text file (or both) in Korean, while also providing the translated copy of the text. Available through the web app or API. Coming soon to the desktop app.

SUMMARY ENGINE – POWERED BY GPT.

Convolutional neural networks build conceptual relationship vectors that automatically weigh and rank document text to determine what the document subject matter is, then uses natural language algorithms to describe the documents content in a way that feels very human. Summary Engine provides an extremely high level of accuracy in it’s summarizations and can be artificially weighted to illuminate certain keywords, concepts or entities where they are encountered. Available through the web app. Coming soon to the desktop app.

DEPOSITION ENGINE – POWERED BY GPT.

Take everything we learned with Summary Engine and completely obsess for a year on curating the best deposition summarization experience imaginable. Delivered in a self-executing HTML wrapper, requiring no web server, internet access or training…. it’s just something you’ll need to experience first hand. We summarize entire depositions in about 20 seconds and the output is searchable, grouped by topic relevance, while thoughtfully placed tooling is built in under each summarized topic when activated, effortlessly unfolds question and answer pairs collected throughout the entire deposition transcript. Going live in the desktop app Q4 2023. Early access is available for qualified participants.

DLP: (DATA LOSS PREVENTION) AS A SERVICE

AI, ML and Natural Language algorithms work to identify strings and patterns of text relative to PII, PHI or anything else using Cullable’s query language or regex. FInd protected information to comply with HIPAA, PCI/DSS, GDPR, CCPA or identify personal or otherwise protected information anywhere. Available through the web app. Coming soon to the desktop app. Enables automated redaction workflows, in case the user needs to obfuscate or ‘scrub’ documents for protected distribution workflows. Popular in FOIA fulfillment. Delivered with a massive library of pre-trained infotype detectors. A current and maintained list of infotype detectors is available here. Custom detectors can also be created.

AUDIO AND VIDEO TO TEXT or TEXT TO SPEECH (TTS)

AI, ML and Natural Language algorithms work to recognize text from audio streams in any language. Can be pipelined into Translation, DLP or Summary Engine. Identifies unique speakers as separate entities. For example, a panel of 8 members in discussion would ideally be recognized as unique individuals throughout the conversation text.

Additionally, we’ve recently announced support for text to speech using studio talent voice that produces an audio stream that is actually indistinguishable from live human recorded text. It’s in our opinion, the highest quality TTS technology available anywhere today. Available on request. Coming soon to the desktop app.

EXTRACT: DOCUMENT AI FORM DATA EXTRACTION.

High accuracy form data extraction as a service. Originally designed to serve medical records form extraction for payment processing, also used in statistical / business intelligence and litigation initiatives. Documents are processed for text and object orientation in order to help train a machine learning model that understands form fields and responses as key value pairs.

Resulting data can be used to populate spreadsheets, databases, CRMs and more. Base technology can be extended to capture other kinds of information, potentially lending itself to categorization or other organizational workflow feats, such as taxonomy definition, ETL, reporting or even discovery initiatives. In development now. Initial, trainable models available Q2/2023.