Neuralk-AI Raises $4 Million to Revolutionize AI for Tabular Data in Retail and Commerce

When we talk about "tabular data," we’re referring to any type of structured information that fits neatly into rows and columns. This could be data from SQL databases, spreadsheets, CSV files, or similar formats. It’s the kind of data businesses have been using for years to track everything from sales numbers to customer behavior.

While AI has made impressive strides in working with unstructured and sequential data, like text and images, these models (especially large language models or LLMs) aren’t perfect when it comes to handling structured, tabular data. LLMs are designed to generate coherent output by manipulating input tokens, but they don’t always follow a fixed, predictable structure. The best of these models can be expensive to use via APIs or to run on cloud infrastructure, which makes them less accessible for smaller businesses.

Despite the advances in AI, many companies already have a solid data infrastructure in place, often in the form of a data warehouse or data lake. They’ve been centralizing important data, and they have teams of data scientists working to leverage this data and drive better business decisions.

Enter Neuralk-AI, a French AI startup focused on applying artificial intelligence to the world of tabular data. This week, the company announced it has raised $4 million in funding to further its efforts.

'For the company to have real data, the data to be used was a long time ago, it was structured in the form of a table, used by data scientists to create machine learning models,' Neuralk-AI co-founder and chief scientific officer Alexandre Pasquiou.

There is a huge need to reimagine the development of AI models, Neuralk-AI focuses on structured, tabular data. Their goal? As the product of a massive engineering effort that started in 2014, we wanted to deploy something to commerce companies—retailers, particularly—that work with on the order of terabytes of access for product catalogs etc.

Pasquiou explains, “LLMs are great for search, natural language interaction, and answering questions based on unstructured documents. But once you return to classic machine learning, which is really built on structured data, LLMs hit some limitations.”

Neuralk-AI’s models aim to help these companies automate complex data workflows, like deduplication and enrichment. But the possibilities go much further—retailers could use these AI tools to spot fraud, optimize product recommendations, and even generate accurate sales forecasts that could assist with inventory management and dynamic pricing.

The startup’s recent $4 million funding round was led by Fly Ventures, with additional participation from Steam AI. Business angels like Thomas Wolf (from Hugging Face), Charles Gorintin (from Alan), and Philippe Corrot and Nagi Letaifa (from Mirakl) also backed the company.

Neuralk-AI’s team is actively working on improving its models, with plans to test them with major French retailers and commerce startups such as E.Leclerc, Auchan, Mirakl, and Lucky Cart.

“We’re aiming to release the first version of our model within three to four months, along with a public benchmark to compare our model with the current industry standards,” said Pasquiou. “By September, we hope to establish ourselves as the best tabular foundation model for representation learning.”