The Company

iKVA is a UK-based start-up founded in 2018 and headquartered in Cambridge, UK. The company positions itself as a knowledge management company that can transform data into business insights. Originally called Kvasir, the firm has developed several technology solutions that leverage machine learning and semantic technologies to, as they put it, “go beyond search.”

The company is led by CEO and co-founder Jon Horden; it has raised approximately $6 million over two funding rounds in 2020 and early 2021. The most recent round came from Cambridge Enterprise, the commercial arm of Cambridge University. Indeed, the company has close ties to the university as the three other co-founders hold Cambridge posts. The iKVA products result from research undertaken there in conjunction with its Computer Lab and the Alan Turing Institute.

The Technology

Six technology solutions fall under the iKVA umbrella, ranging from Intelligent Insights for Lawyers to a bid-building tool (see Figure 1). Underlying all these tools is a shared system and approach. At the core is the use of “vector mapping” that the company utilizes over more traditional natural language processing (NLP) approaches to text and data analysis. Vector mapping provides insights from unstructured data to enable knowledge workers to do their jobs. Put simply, vector mapping converts text into numerical values and spaces; this means that rather than examining a sequence of words, it analyzes a series of points (vectors) within a semantic space. In practical terms, that means you can analyze any text, in any format, in any language, whether within a document or a conversation.

Further, unlike more traditional approaches, the iKVA system does not rely on good quality metadata, tagging, or taxonomies. Of course this is not magic, and a lot of work is needed to pre-train the system before using it. iKVA has been pre-trained for multiple different use cases (hence the six solutions now available) and will be trained for others in the future. Typically, specific customer training is needed to optimize the system, but all in all it’s pretty much ready to run out of the box.

Though there is much complexity beneath the covers, architecturally iKVA runs in a distributed cloud environment (typically AWS or Azure). Ideally, a customer will opt to run on a dedicated virtual server within the iKVA environment. Within this private instance, you would run the indexer and the APIs to connect you to your existing data silos and archives. In parallel, pre-trained “knowledge packs” that contain the intelligence of iKVA will process and parse incoming data. UIs (user interfaces) are designed to meet specific customer requirements and run either standalone or embedded within existing working applications such as Microsoft Office, Slack, etc. Another option is to run iKVA as a private cloud. To ensure that the system is secure, all communications run encrypted over HTTPS, and the infrastructure itself is certified to ISO 27001.2013 standards. So, this is a pretty straightforward computing environment. And as with any such system, the time taken to install it will largely be consumed in finding and creating connections to all the relevant data sources.

The secret sauce for iKVA is contained within and through the creation of the knowledge packs. In conjunction with the indexing capabilities, iKVA processes unstructured data from all the connected data sources through neural networks (vector mapping) to create a single unified index without the need for extensive tagging or taxonomy building. In practice, this means that data points are in one place where they can be graphically analyzed to identify clusters and previously hidden relationships matching your specific task or corporate needs. (This differs from a traditional search approach that relies on keyword queries running across federated sources.)

In this manner, the system can a) understand the structure, content, and context of the data regardless of format or language and enable users to find and make use of disparate information sources without needing keywords, or, ideally, b) have relevant insights surfaced to the user automatically. For example, users of the system can automatically receive or define insights that are guided by past query activities, or that are matched to pre-defined corporate requirements. Similarly, relevant insights can be surfaced within the user’s environment (for example Microsoft Word) based upon their current work activity.

The sweet spot for iKVA is its potential use within organizations that rely heavily on accessing and making sense of large volumes of disparate data/information, and, equally importantly, those that would benefit from automating insight generation from those volumes.