Deasie, a startup that aims to provide companies with better control over text-generating AI models, has successfully raised $2.9 million in a seed funding round. The funding was led by prominent investors, including Y Combinator, General Catalyst, RTP Global, Rebel Fund, and J12 Ventures.
Key Takeaway
Deasie has raised $2.9 million in funding to enhance the reliability of generative AI by ranking and filtering data. By connecting to unstructured company data, Deasie’s platform automatically categorizes and evaluates the data, enabling the use of relevant, high-quality, and safe information in text-generating models.
The Need for Improved Governance in Generative AI
The founders of Deasie, Reece Griffiths, Mikko Peiponen, and Leo Platzer, have extensive experience in developing data governance tools during their tenure at McKinsey. They discovered significant challenges and opportunities related to enterprise data governance that could potentially hinder the adoption and effectiveness of generative AI.
A recent survey conducted by IDC revealed that 86% of executives at large enterprises believe that more governance is required to ensure the quality and integrity of generative AI insights. Surprisingly, only 30% of the participants considered themselves adequately prepared to leverage generative AI in their organizations.
Making Generative AI Models More Reliable
With the aim of improving the reliability of generative AI models, Deasie has developed a product that connects to unstructured company data, such as documents, reports, and emails. This product automatically categorizes the data based on its contents and sensitivity.
For example, Deasie can tag a report as “personally identifiable information” or “proprietary information” and indicate its version number. It can also label a spec sheet as “proprietary information” with restricted access rights. Clients of Deasie define the tags and labels to customize the classification of their data, thus training Deasie’s algorithms to classify future data accurately.
After auto-tagging the documents, Deasie evaluates the corresponding data in terms of relevance and importance. Based on this assessment, Deasie decides which data should be fed into a text-generating model.
Ensuring High-Quality and Safe Data for Generative AI
Reece Griffiths, one of the founders, explains that enterprises possess vast amounts of unstructured data that often lack proper governance. The larger the volume of data, the higher the chances that language models will generate irrelevant or sensitive information. Deasie’s platform filters through an enterprise’s documents to ensure that the data used in generative AI applications is relevant, of high quality, and safe.
Although Deasie’s platform is intriguing, questions arise regarding the consistency of their data classification algorithms and the accuracy of assessing a document’s importance. However, Deasie appears to have gained traction in the market, with a significant pipeline of enterprise customers, including a multi-billion-dollar enterprise in the United States and five Fortune 500 companies.
In the coming months, Deasie plans to expand its engineering team and introduce new features to differentiate itself from competitors like Unstructured.io, Scale AI, Collibra, and Alation.
