Knowledge
Augmented Document Assistant
Consumer-facing versions of generative AI (GenAI) are familiar to many, if not most, people nowadays. They may not always be quite sure how it works, but they know what it can do, whether that be drafting a career resumé or an email to a local newspaper.
But technologies such as this only go so far. They tend to be generic, in terms of their output, of their functionality, and of the data sources on which they draw. To be useful in more commercial or technical engineering environments, what’s needed is document processing of greater breadth, depth, speed, and specificity.
Fit for purpose
That’s what Capgemini Engineering’s Augmented Document Assistant sets out to provide. It’s an AI asset that uses an open-source approach to analyze and extract information from documents as well as to author new ones. What’s more, it does this in ways that suit the particular environment in which it is operating, drawing on relevant rather than
broad data sources, and using the terminology, comprehension levels, formats, and language appropriate to the given context.
Documents convey information in different ways: text, images and tables of data all have meaning. Augmented Document Assistant absorbs information at scale from all these in PDF, Word, PowerPoint, and similar unstructured formats, and uses open-source Natural Language Processing (NLP) and Large Language Models that are applicable to circumstances to provide accurate and nuanced answers. Those answers are auditable, so users can easily go to the original documents to see the source of Augmented Document Assistant's insights.
The solution is highly customizable. Its architecture can be amended and extended, allowing organizations to create and integrate the best models for their requirements – and because Augmented Document Assistant is not API based, it also works offline. It’s platform agnostic: because it’s based on open-source state-of-the-art models, it’s compatible with any public or private cloud, or other on-premise environments. In addition, it seamlessly inherits the security policies of its client environment.
Comprehensive and customizable
While some AI document analysis solutions need a full end-to-end stack to be created, with Augmented Document Assistant, everything is fully tested and already in place. The chat function is joined by facilities including a search engine, authoring, name entity recognition, summarization, and multi-language translation, all of which act upon client corporate data and other sources, and all of which can be customized and fine-tuned by Capgemini to meet the needs of individual organizations. Conversational question-and-answer interactions provide a natural user experience, and the solution facilitates collaboration and information sharing.
Authoring features include the potential to create a digital knowledge base; automatic text suggestions and auto-compilation; the automatic creation of template-based documents; proofreading, including grammar correction and text readability; layout suggestions for copy elements; the automated creation of tables of contents, tables of figures, and image captions, and even more advanced textual comparisons.
Augmented Document Assistant in action
The Augmented Document Assistant approach can help organizations save time and money by streamlining the retrieval of information from documents and by boosting productivity while authoring new documents.
For instance, an aerospace and defense company needed an assistant to support pilots and technicians in the retrieval of information from helicopter manuals, which contain more than 30,000 pages for each helicopter type. Augmented Document Assistant created a digital knowledge base of pre-processed documentation that could be navigated interactively, enabling users to locate information across multiple documents and pages with its search engine and to interrogate the contents. The tool also describes technical tables and images, thus unlocking the complex information they contain. All information can be accessed in multiple languages and can be instantaneously summarized. Text-to-speech and speech-to-text implementations mean people can interact with Augmented Document Assistant orally, instead of simply by reading and writing text.
Similarly, an oil and gas company wanted to develop an AI virtual assistant capable of processing over 20,000 geoscience reports and other material and extracting useful suggestions to assist in the editing of live documents. Augmented Document Assistant used machine learning and advanced NLP models and methods to extract, process and score its own suggestions, writing PDF documents of a specific format and structure. In particular, the solution provided textual suggestions that facilitate the compilation of new PDF documents using the data previously extracted from the archive of documents of the same type. These suggestions were related to entities that were important to the Geoscience Department, such as wells and their specific types.
Fulfilling a global need
In a world that is creating information at a blinding pace, the need for technology that can assess it all and help to derive insight from it is greater than ever. Augmented Document Assistant has been designed to meet that need.
Augmented Document Assistant – what’s involved:
Highly configurable – state-of-the-art open-source AI which is not a black box, but accessible and customizable at any time in a specific domain
Highly scalable – can handle large volumes of textual data in different formats with a high level of technical detail
Flexible IT environment – can be integrated with any type of private or public cloud or on-premise IT environment, working both online or offline
Development efficiency – saves more than a year of technical developments starting from the set of functionalities already in place
Business productivity – reduces information retrieval time from days to seconds and the authoring time from weeks to days
