Skip to main content

Neuralk AI Documentation

Neuralk AI is a deeptech startup building powerful foundation models for machine learning (ML) tasks across industries like commerce, finance, healthcare and more.

The Neuralk API empowers data science and ML teams with state-of-the-art prediction models and expert AI modules that are ready to be deployed out of the box for use cases found in the above industries.

Get Started

Neuralk AI Prediction Models​

Tabular Foundation Models​

Tabular Foundation Models (TFMs) are a new generation of frontier AI models capable of making predictions on tabular data for a variety of ML tasks (classification, regression, time-series) instantly and without any additional model training.

At Neuralk AI we develop proprietary TFMs for industry-grade applications that deliver predictions at a state-of-the-art accuracy (explore our benchmark here).

NICL (Neuralk In-Context Learning)​

NICL is a tabular foundation model developed by Neuralk AI that leverages in-context learning to deliver instant & highly accurate predictions with only a small set of examples provided as context.

NICL performs best on datasets with up to 250 features and 15,000 samples, and can can scale up to 1 million samples and 500 features.

NICL is ideal when you:

  • Need strong baseline performance without hyper-parameter tuning.
  • Want a unified approach to handle mixed feature types.
  • Are exploring new datasets and want fast iteration.
  • Prefer interpretability and flexible prompting over black-box optimisation.

Read more on its usage here.

Expert Neuralk AI Modules​

The Neuralk API gives you access to Expert AI modules: prebuilt, industry-ready workflows ready to solve complex business use cases out of the box.

This solution lets data science and ML teams skip building complex pipelines from scratch, minimizing engineering effort and delivering high-precision insights to business teams with minimal delays (see our benchmark here).

End-to-end workflows​

These modules are called “expert” because they go beyond the delivery of predictions: they automate the entire workflow of a predictive use case from start to finish, including:

  • Data preprocessing: Transforming messy tabular data into clean, model-ready inputs, tailored to the target use case.
  • Feature engineering: Extracting and refining meaningful features that serve as inputs to the model and can boost predictive performance.
  • Use case–specific logic: Applying domain-specific rules and additional preprocessing steps that help boost model performance even further.
  • Seamless prediction: Integrating seamlessly NICL into your workflow to generate high-accuracy predictions instantly, without any additional model training.

The end-to-end modules provided are fully optimized for each use case, ensuring every step delivers fast, precise results with minimal engineering overhead.

Example of use cases include product categorization (commerce), fraud detection (finance), churn prediction and more. Discover all of our use cases on our website.

On this page