Fundamental Launches NEXUS to Transform Business Data into Predictions

The artificial intelligence company Fundamental, based in San Francisco, announces the launch of NEXUS, a machine learning model designed to process tabular data. With $255 million in funding, NEXUS promises to revolutionize the way businesses handle forecasts, eliminating the need for traditional manual feature engineering processes.

NEXUS stands out for understanding business data as a complex network of relationships, rather than a simple sequence. Therefore, it can identify patterns that often go unnoticed by conventional methods.

The technology: overcoming sequential logic

Many current AI models operate on sequential logic, predicting the next word or pixel. However, business data is not sequential but rather multidimensional. For example, predicting customer churn risk involves an intersection of transaction frequency and sentiment analysis.

Jeremy Fraenkel, CEO of Fundamental, states: "The most valuable data is in tables, and until now there hasn't been a suitable foundational model to understand them." He emphasizes that companies still rely on traditional algorithms that do not leverage the richness of tabular data.

NEXUS was trained on billions of tabular datasets using the Amazon SageMaker HyperPod platform. Instead of requiring data scientists to manually define features, NEXUS is capable of ingesting raw tables and identifying hidden patterns.

Tokenization challenges

One of the reasons why LLMs fail to handle tabular data is how they process numbers. Fraenkel explains that LLMs fragment numbers into tokens, similar to how they do with words. This results in a loss of understanding of numerical distribution.

NEXUS, on the other hand, is designed to recognize that a change in column position should not affect the prediction. It understands that a table should be seen in its entirety.

Operating on the predictive layer

Unlike integrations like Anthropic's Claude with Excel, NEXUS operates on a predictive layer rather than a formula. It is used for quick decisions, such as determining fraudulent transactions in real-time.

While tools like Claude summarize tables, NEXUS predicts outcomes, whether it’s a factory failure or the probability of hospital readmission.

Architecture and availability

Fundamental claims that NEXUS significantly reduces the time required to gain insights. The process of building a predictive model that once took months can now be accomplished with just one line of code.

As part of its market strategy, Fundamental has already secured seven-figure contracts with Fortune 100 companies, utilizing Amazon Web Services (AWS) as a vendor. This facilitates technology adoption, allowing business leaders to use existing AWS credits.

Social implications of the technology

Beyond commercial benefits, Fundamental emphasizes the social importance of its technology. NEXUS can predict significant failures in critical sectors, such as health and infrastructure.

NEXUS can identify hidden signals in structured data, preventing crises, such as the Flint water crisis, which resulted in billion-dollar damages.

Performance versus latency

What is considered better varies by sector. In sectors like health, the focus may be on accuracy, while in finance, even a slight increase in accuracy can yield significant gains. Fraenkel states: "A small increase in accuracy can be worth billions to a bank."

Ambitious vision and significant support

The $225 million investment led by Oak HC/FT demonstrates confidence in the potential of tabular data. Annie Lamont from Oak HC/FT stated: "The importance of Fundamental's model is undeniable."

Strategic partnership with AWS

Fundamental has partnered with AWS, allowing NEXUS to be integrated directly into the AWS dashboard. The solution keeps data secure, being the first company to offer fully encrypted models within the client's environment.

NEXUS presents a challenge to the traditional way of business decision-making, offering a new era of predictive intelligence that can transform how operational models are conceived. If NEXUS performs as promised, it could be the first step towards AI truly learning to read the data that underpins the global economy.