Multilingual NLU

HumanFirst's natural language exploration and labeling features can be powered by our native multilingual NLU model, trained on-demand on your labeled data.

API reference

Model assisted active learning labeling workflows

HumanFirst allows you to quickly drill down into your unlabeled data to find semantically similar utterances, and provides powerful workflows to label hundreds of utterances in a single click.

Guide Try demo

Full-text and semantic search

HumanFirst indexes and prepares your entire unlabeled data for full-text search as well as semantic search. Our semantic search feature has two modes:  real-time KNN, or using the trained NLU model across your intents.

Documentation

Interactive real-time clustering

Explore your data with interactive clustering,  quickly modify the clustering parameters (granularity, size of clusters) in real-time.

Documentation

Actionable coverage analysis

Estimate the coverage of your existing intents within unlabeled data, with interactive thresholds that allow you to adapt results based on the model's performance.

Documentation coming soon

Use your intent hierarchy to discover other intents

Sort and explore your unlabeled data based on  uncertainty, margin score and entropy report of your trained model, to identify utterances that are most likely to represent new or related intents.

Documentation

Flexible, hierarchical intent organization

Easily scale and manage up to thousands of intents, with hierarchical organization of intents.
Import flat intents from your existing NLU platforms, and automatically transform them into easy-to-explore hierarchies: HumanFirst allows you to specify the delimiter used in your intent names, and flattens out your intents when you export the data.

Documentation Guide

Re-usable intent and entity data across workspaces

Build and maintain modular intent catalogs that can be re-used across verticals and projects.
Import and re-use any intent (and its accompanying training data) across workspaces.

Documentation coming soon

Split and merge intents

HumanFirst provides machine-learning assisted workflows that help fix your model's data: merge conflicting intents and their training phrases together with a single click, or quickly move problematic utterances from one intent to the other.

Guide

Intent disambiguation workflows

sHumanFirst's built-in disambiguation feature (both real-time, and based on your trained model) allows you to quickly view what intents are conflicting, and provides actionable workflows to ensure each intent's scope is as clean and specific as possible.

Documentation coming soon

Entity discovery and management

Label and organize your entities and synonyms, and explore your unlabeled data using these as search keywords.

Documentation coming soon

Export intent and entity data in any format

HumanFirst allows you to export your work back in a number of formats, ready for use within your project. You can do this manually, through our APIs or via our command-line for your entire labeled data, or selectively for a subset of your intents/entities.

API reference

Save additional tags and metadata on any intent or training phrase

Keep your data in sync between HumanFirst and your other tools: tags or metadata added to your intents or training phrases will be maintained during import and export, and can be created/modified within HumanFirst, providing a very complete and flexible meta-store.

Documentation coming soon

K-fold cross-validation

HumanFirst generates on-demand 5-fold cross-validation analysis against its NLU (or your own), to provide intent-level metrics (F1, precision, recall, accuracy) that can be used to understand and tune your model.

Documentation coming soon

Exportable confusion matrix

You can export the raw confusion matrix resulting from the cross-validation analysis for offline analysis, or to apply within your existing workflow.

Documentation coming soon