UiPath used a virtual AI Summit hosted in March to showcase a range of new generative AI (gen AI) features in its platform that the company says will help organizations realize the full potential of AI for automation with specialized AI models tailored to their industry challenges and use cases.
Graham Sheldon, Chief Product Officer at UiPath, said that UiPath’s new large language models (LLMs) and a Context Grounding feature will enable companies to optimize GenAI models with business-specific data. This ensures that AI has the integrations, data, context, and ability to automate processes in line with customers’ unique needs.
UiPath has, over recent years, invested in gen AI as part of an effort to build an end-to-end automation and Intelligent Document Processing (IDP) platform. The 11 new features pair generative AI with specialized AI to reduce training and validation time, providing faster time to value.
The new LLMs, DocPATH and CommPATH, give businesses LLMs that are extensively trained for their specific tasks, document processing and communications. General-purpose GenAI models like GPT-4 struggle to match the performance and accuracy of models specially trained for a specific task.
Instead of relying on imprecise and time-consuming prompt engineering, DocPATH and CommPATH provide businesses with extensive tools to customize AI models to their exact requirements, allowing them to understand any document and a huge variety of message types.
DocPATH is a generative LLM specialized for a large and diverse set of enterprise documents and will go into public preview in the second quarter of the year. UiPath claims it needs little to no training time, with up to 10x training effort reduction and many use cases requiring no annotation.
In addition, it allows for more accurate document extraction, with a 45% to 76% error rate reduction compared to market leading generative AI models. It can process complex tables: 30% to 65% less errors compared to other IDP and Generative AI vendors, according to UiPath.
CommPATH is a generative LLM built and trained for communications data. This LLM increases the automation rate, extracts complex data, and understands the requests, fields and the relations between them. UiPath says it needs two to three times less training effort for the same levels of accuracy compared to conventional Entity training.
Augmenting GenAI models with business specific data
Context Grounding, a new feature within the UiPath AI Trust Layer that entered private preview in April, helps businesses improve the accuracy of GenAI models by providing a foundation of business context through retrieval augmented generation. This system extracts information from company specific datasets, like a knowledge base or internal policies and procedures to create more accurate and insightful responses.
Context Grounding makes business data LLM-ready by converting it to an optimized format that can easily be indexed, searched, and injected into prompts to improve GenAI predictions. Context Grounding will enhance gen AI experiences in UiPath Autopilots, GenAI Activities, and UiPath IDP products.
UiPath also announced that IBM has used UiPath Connector Builder to create a unique watsonx.ai connector. The new connector provides UiPath customers with access to multiple foundational models available in watsonx.ai. Use cases, such as summarization, Q&A, task classification, and optimization for chat, are infused into UiPath workflows.
IBM watsonx customers can also access broader UiPath platform capabilities, such as Test Automation, Process Mining and Studio workflows, all within a low/no-code UX environment. IBM’s consulting capabilities, coupled with the UiPath Business Automation Platform, will help support successful GenAI adoption, including the right strategy for infusing AI into more powerful, and complex automated workflows.
Autopilot for developers & testers
Also announced alongside the AI Summit, UiPath Autopilot is a suite of GenAI-powered experiences across the platform that make automation builders and users more productive. Autopilot experiences for developers and testers are available in preview with a targeted general availability in June.
Autopilot for Developers empowers both professional and citizen automation developers to create automations, code, and expressions with natural language, accelerating every aspect of building automations. Autopilot for Testers transforms the testing lifecycle, from planning to analysis, reducing the burden of manual testing.
New prebuilt Gen AI Activities, meanwhile, use the UiPath AI Trust Layer and are easy to access, develop with, and leverage high quality AI predictions in automation workflows that deliver faster time to value. Activities provide access to use cases, such as text completion for emails, categorization, image detection and language translation.
