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 enterprises put AI models to work immediately, with full confidence that they won’t experience any problems with hallucinations and bias. To do this, Kolena has created a number of specialized AI agents that can perform complex tasks such as document understanding, comparison and analysis in order to boost the performance of AI models.

Founded in 2021, Kolena first developed an AI and machine learning model testing platform that allows teams to do rigorous tests on the behavior of their algorithms. With its platform, enterprises can assemble highly specific test cases from the data sets that power their AI models, enabling evaluation of each model’s effectiveness in key scenarios and applications.

With Restructured, Kolena says, it’s going further, helping enterprises to leave “outdated” techniques such as retrieval-augmented generation behind. Instead, it combines RAG with vector search, natural language processing models and multimodal large language models, so companies can extract actionable insights from data such as documents, contracts, audio files, video scans, images or just about any other source of information. With this, it can help to ensure the utmost accuracy with enterprises’ AI applications.

Restructured is important because it enables enterprises to supplement their AI algorithms’ knowledge with unstructured data, which according to Gartner Inc. accounts for about 80% to 90% of the total data amassed by the average organization.

Kolena’s co-founder and Chief Executive Mohamed Elgendy said Restructured introduces AI agents that enable engineers to harness insights from previously isolated data on an unprecedented scale. This can help to reduce hallucinations, when AI models simply fabricate responses to questions they’re unable to answer.

“We’re not building just another chat tool; Restructured is a productivity platform for real work – worksheets, project plans, data digitization, document search and understanding, analytics dashboards, and more,” Elgendy said. “If you’re a professional dealing with vast amounts of data, where accuracy and speed matter, we’re building workflows for you.”

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

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

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

Constellation Research Inc. Principal 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 around intelligent document search and processing.

“Kolena was primarily a data quality company, and for a long while it was focused on testing and enhancing model quality, so it’s interesting to see it branch out into agentic AI,” Thurai said. “It’s looking to cater to customers that have integrated generative AI solutions such as call center, customer service chatbots and ITOps and observability solutions.”

The analyst said its dynamic, agent-based search sounds like a compelling solution, but he wondered if it’s strong enough to succeed as a standalone product. “It might be interesting for some legacy vendors in the above areas that struggle to invest fast enough,” he said.

The startup adds that its platform has already been put to use by a number of early adopters, helping them to reduce the time they spend on legal research, data discovery and due diligence. It cites the example of a “top-tier law firm” that accelerated its case discovery processes by up to 80% using its AI agents.

In addition, insurance firms are using 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 to do using traditional search methods.

Image: SiliconANGLE/Microsoft Designer