Artificial intelligence startup Kolena Inc. is moving beyond AI model testing and quality control with the launch of its new flagship platform, Restructured.

The company says its goal is to help companies implement their AI models immediately, confident that they won’t encounter any issues with hallucinations and biases. To do this, Kolena has created a number of specialized AI agents capable of performing complex tasks such as understanding, comparing and analyzing documents to improve the performance of AI models.

Founded in 2021, Kolena first developed a AI and machine learning model testing platform which allows teams to conduct rigorous tests on the behavior of their algorithms. Through its platform, companies can assemble highly specific test cases from the data sets that power their AI models, allowing the effectiveness of each model to be evaluated in key scenarios and applications.

With Restructured, Kolena says, it goes further, helping companies move away from “outdated” techniques such as recovery-augmented generation. Instead, it combines RAG with vector search, natural language processing models, and multimodal extended language models, so businesses can extract actionable insights from data like documents, contracts, audio files, video analytics, images or just about any other source of information. Thus, this can help ensure the highest accuracy with enterprise AI applications.

The restructuring is important because it allows companies to supplement the insights of their AI algorithms with unstructured data, which Gartner Inc. estimates accounts for about 80 to 90 percent of the total data accumulated by an average organization.

Mohamed Elgendy, co-founder and chief executive of Kolena, said Restructured introduces AI agents that enable engineers to harness insights from previously isolated data on an unprecedented scale. This can help reduce hallucinations, when AI models simply fabricate answers to questions they are unable to answer.

“We’re not just building another chat tool; Restructured is a productivity platform for real work: worksheets, project plans, data scanning, document searching and understanding, analytics dashboards, and more. said Elgendy. “If you are a professional dealing with large amounts of data, where accuracy and speed are important, we build workflows for you. »

The key element of Restructured is Kolena’s proprietary AI Task Planner agent, which structures all the most relevant data for a given AI model or application into a single data stream. The model can then use Kolena’s advanced search capabilities to find the right data to accurately answer any request or question. It goes further than just extracting information, as it can also apply numerical and mathematical equations to the data it retrieves.

Additionally, Restructured provides tools for creating real-time analytics dashboards, so users can easily monitor the performance of their generative AI applications. Dashboards also ensure full transparency, mapping the data sources and analytical steps its AI agents use and perform, so users have a way to verify accuracy.

The platform can be deployed either as a traditional software-as-a-service offering or directly on-premises for businesses that need to ensure sensitive data remains locked on their own servers. According to Kolena, the restructuring will have the greatest impact in sectors such as finance, law, real estate, insurance, research, knowledge management and business training.

Constellation Research Inc. Senior Analyst Andy Thurai told SiliconANGLE that today’s announcement shows how Kolena is diversifying its core business, using AI agents to help with key tasks related to intelligent document searching and processing.

“Kolena was primarily a data quality company and for a long time focused on testing and improving model quality. So it’s interesting to see her move into agentic AI,” Thurai said. “It seeks to cater to customers who have integrated generative AI solutions such as call centers, customer service chatbots, and ITOps and observability solutions. »

The analyst said its dynamic agent-based search appears to be a compelling solution, but he questions whether it is powerful enough to succeed as a standalone product. “This could be interesting for some incumbent providers in the above areas who are struggling to invest quickly enough,” he said.

The startup adds that its platform has already been used by a number of early adopters, helping them reduce the time spent on legal research, data discovery and due diligence. He cites the example of a “leading law firm” that accelerated its case discovery processes by up to 80% thanks to its AI agents.

Additionally, insurance companies use Restructured to sort through millions of pages of policies, regulations and contracts to uncover specific clauses, precedents and risk factors. Such tasks previously took weeks with traditional research methods.

Image: SiliconANGLE/Microsoft Designer