The term “tabular data” refers to structured datasets typically organized in rows and columns, such as those found in SQL databases, spreadsheets, or .CSV files.
Although significant advancements have been made in artificial intelligence concerning unstructured and sequential data, large language models (LLMs) inherently prioritize flexibility. These models are designed to process input tokens to produce coherent output without adhering to a specific structure. Accessing the top-tier LLMs can be costly, whether through APIs or by operating them on private cloud systems.
Nevertheless, numerous organizations have established data strategies that include a centralized data warehouse or lake and have employed data scientists to harness this information for enhancing corporate strategies.
The French startup Neuralk-AI focuses on artificial intelligence models tailored for tabular data and recently secured $4 million in funding.
“The data that holds real significance for businesses is that which has long been recognized, organized into tables, and utilized by their data scientists to develop machine learning algorithms,” stated Alexandre Pasquiou, co-founder and chief scientific officer of Neuralk-AI, in an interview with TechCrunch.
Neuralk-AI sees a promising opportunity to reshape AI model development with an emphasis on structured data. Initially, the company aims to provide its model as an API directed at data scientists in the retail sector, where data is highly valued—encompassing elements like product catalogs, customer information, and trends in shopping carts.
“Currently, LLMs excel in functions like search, human-like interaction, and answering queries based on unstructured documents. However, they face challenges when we revert to traditional machine learning, which fundamentally relies on classical tabular data,” Pasquiou explained.
With Neuralk-AI’s offerings, retailers can streamline intricate data workflows through intelligent deduplication and data enrichment. Moreover, these models can aid in fraud detection, enhance product recommendations, and create sales forecasts that inform inventory and pricing strategies.
The $4 million funding round was led by Fly Ventures, with participation from Steam AI. Additional investments came from notable angel investors including Thomas Wolf from Hugging Face, Charles Gorintin from Alan, and Philippe Corrot and Nagi Letaifa from Mirakl.
The team is diligently refining its models and plans to conduct testing with prominent French retailers and commerce startups, such as E.Leclerc, Auchan, Mirakl, and Lucky Cart.
“In about three to four months, we will unveil the preliminary version of our model along with a public benchmark that will allow us to assess its performance against the leading solutions in the field,” Pasquiou stated. “By September, our objective is to establish ourselves as the foremost tabular foundation model in matters related to representation learning.”
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
