Tabular data are a broad term that covers structured data that generally conforms to a specific row and column. It can be a SQL database, a spreadsheet, a .CSV file, etc.
While there has been great progress in artificial intelligence applied to unstructured and sequencial data, these large -language models are confused by design. They are built to manipulate input tokens to generate a coherent output without necessarily following a fixed structure. The best LLM are also expensive to access through an API or expensive to execute in their own cloud infrastructure.
And yet, many companies already have a data strategy with a data warehouse or a data lake to centralize all important data and some data scientists who can take advantage of these data to improve the company’s strategy.
French startup Neuralk-Ai It is an artificial intelligence company that has been working on models of ia focused on tabular data. The company announced $ 4 million this week in funds.
“The data with real value for the companies are data that were identified a long time ago, structured in the form of a table and used by the data scientists of these companies to create all their automatic learning algorithms”, co-founder of Nueralk-AI and the Scientific Director Alexandre Pasquiou told Techcrunch.
Neuralk-Ai believes that there is an opportunity to review the development of the AI model, but with a specific approach to structured data. At first, he plans to offer his model as API to data scientists who work for commerce companies because these companies love Data: Think of product catalogs, customer databases, shopping cart trends, etc.
“Today, the LLMs are excellent for the search, the user’s natural interaction and answer questions based on unstructured documents. But it has some limitations at the time we return to classic automatic learning, which is really based on classic tabular data, “said Pasquiou.
With Neuralk-AI, retailers can automate complex data workflows with intelligent deduplication and enrichment. But the company’s models could also use to detect fraud, optimize product recommendations and generate sales forecasts that could be used for inventory management and product prices.
Fly Ventures led the company’s round of the company with Steamai also participating. Several business angels also invested in the startup, such as Thomas Wolf of Hugging Face, Charles Gorintin by Alan and Philippe Corrot and Nagi Lartaifa de Mirakl.
The team is still actively working on its models. He plans to try a group of retailers and new leading French companies, such as E.leclerc, Auchan, Mirakl and Lucky Cart.
“Within three or four months, we will launch the first version of our model and the public reference point in which we can classify our model compared to the state of art in this space,” said Pasquiou. “And in September, the idea is to be the best tabular base model in everything related to representation learning.”