Anote - Human Centered AI

Organizations within the DOD generate millions of documents (PDFs, PPTX, DOCX, CSV, XLSX, EMAILS) per month that need to be tagged for Controlled Unclassified Information (CUI) content. These documents contain sensitive information, so it is crucial to tag these documents with correct CUI categories to prevent highly classified documents from falling into the wrong peopleĺs hands. However, tagging CUI accurately is very difficult, as the CUI labels correspond to over 100 CUI classifications sitting in the CUI Registry, with classes that include PRVCY, HLTH, and LEI. Knowing how to properly tag CUI data requires subject matter expertise, and involves having a strong understanding of the rules designated in the CUI Registry.áTagging documents with the correct CUI label is mission critical, as CUI tags correlate directly to distribution and dissemination of information. If documents are under classified, people with not appropriate privilege levels have access to information that they are not allowed to access, which is a major security risk. If documents are over classified, DOD personnel that should be able to collaborate and work on solving critical national security problems are not able to have access, which causes inefficiencies in productivity, communications, and collaboration.In order to build a robust CUI tagging solution, there are many complex requirements that need to be met. áThe solution needs easy to use, able to be run on privately on premise, provide accurate predictions with probability scores to measure certainty, have the CUI predictions be recomposed into documents across various file formats, incorporate dynamic taxonomies for evolving CUI categories, and enable SME-involved active learning for enhanced model precision. Anote's CUI tagging solution meets these requirements, and has the following components:Anote - An end to end MLOps platform that enables you to obtain the best large language model for tagging CUI. Anote provides an evaluation framework to compare zero shot LLMs like GPT, Claude, Llama3 and Mistral, with fine tuned LLMs that are trained on domain specific training data (via supervised, unsupervised and RLHF fine tuning). Anote provides a data annotation interface to convert raw unstructured data into an LLM ready format, and incorporate subject matter expertise into your training process to improve model accuracies.Private Chatbot - A private, on premise, chat interface where you could upload documents and download the recomposed CUI markings.áUpreach - An email system such that when you would send the email, determines whether uploaded documents and content have specific categories of CUI, and cross-reference the associated privilege level of email recipients to ensure that they are supposed to receive documents.Software Development Kit - An integration layer between Anote's fine tuned LLMs and dissemination systems within the DOD to plug in CUI tagging models into your system.