Behind every graph, the data: Adaption presents Visualize

Adaption launches Visualize, an AI-powered data analytics tool that transforms questions into interactive charts and allows users to examine their sources.

A chart can reveal a trend, but it doesn't always show the data behind it. With Visualize, Adaption offers an interface to query a dataset in natural language, explore the results as interactive visualizations, and trace back to the exact rows behind every observation. Available now, the tool debuts Adaptive Interfaces, a new component of the ecosystem developed by the company to improve artificial intelligence models based on their training data. Querying your data without manually building charts

Visualize supports both textual and tabular data. Users can import a dataset, ask a question, and receive an analysis along with an interactive chart.

Rather than manually configuring axes, filters, and categories, they can ask which topics are most represented, how certain behaviors vary across multiple groups, or what relationships emerge within the available information.

The interface also suggests questions tailored to the dataset's content. After each analysis, new prompts allow users to delve deeper into a result or explore another dimension.

Successive questions take previous answers into account. An exploration can thus begin with a broad comparison and then gradually focus on a specific category, time frame, or subset.

This approach is particularly aimed at teams working on data intended for training or evaluating AI models, without necessarily wanting to build queries or dashboards for every new hypothesis. Tracing from the chart back to the original data

One of Visualize's core features is connecting visual results back to the source data they originate from.

A user can select a portion of a chart to view the corresponding records. This makes it possible to go beyond simple statistical observation and examine the individual examples that make up a category.

Adaption illustrates this method using a customer support dataset. Each row contains a received message, a provided response, an assigned department, a priority level, and a language.

A team developing an assistant capable of responding in English and German could ask how request types vary between these two languages.

Visualize generates a representation of the results and then allows for further analysis: comparing priorities, examining the departments involved, or studying differences in response phrasing.

Users can also open the original messages and associated responses to compare the observed trends against the actual content of the conversations. Selecting a trend to examine it more precisely

The Highlight feature allows users to isolate a section of a visualization to focus the analysis on the corresponding data.

In the customer support example, a chart might show a higher proportion of resolved refunds among requests written in German.

The user can select this category and ask which departments handled the requests, how the responses were phrased, or whether priority levels differ.

Visualize then narrows the exploration to the selected subset and offers follow-up questions.

This interaction enables a smooth transition from an aggregated observation to the examination of specific records without recreating a complete analysis.

The example provided by Adaption is illustrative. It does not constitute a verified statistic on the performance of German or English-speaking customer service departments. Modifying charts through instructions

Visualize goes beyond automatically generating charts; it also allows users to modify their presentation.

Users can request a different sorting order, highlight a category, or change specific visual elements simply by describing the desired transformation.

The original and modified versions can be displayed side-by-side to compare results. A version can then be saved and shared with a team.

This feature separates the work done on the data from the work dedicated to its presentation. Users can preserve an analysis while adapting its display to the information they wish to examine or communicate.

However, Adaption's announcement does not detail an exhaustive list of chart types, available export formats, or dataset volume limits. Examining data before training a model

Visualize is part of a broader endeavor than merely generating charts.

Adaption aims to help teams understand what a model is likely to learn from its data.

For instance, a training set might overrepresent certain requests, overlook a category of issues, or contain responses whose style varies by language.

These differences can influence the behavior of a specialized assistant. Visualize allows teams to examine them before deciding whether to keep, modify, or supplement specific examples.

In the case of customer support, a team might identify more detailed responses in one language than in another, or disparities in handling across request categories.

The visualization then serves to guide decisions regarding the data, rather than directly generating a new version of the model. Three components to adapt AI systems

Visualize is the first application of Adaptive Interfaces, the third component of the Adaption platform.

The company is already developing two other technology suites.

Adaptive Data enables the creation, transformation, and organization of data used to train custom models.

AutoScientist, integrated into Adaptive Intelligence, automates part of the work involved in optimizing models and their training methods, taking the available data into account.

Adaptive Interfaces operates upstream and alongside this process, giving users the tools to examine data and specify the behaviors they wish to achieve.

The three components therefore serve distinct functions: Visualize allows for information exploration, Adaptive Data is used to adapt it, and AutoScientist handles the training phase.

Adaption's stated objective is to build custom models that evolve based on user feedback and observations. A first tool accessible from the Adaption platform

Visualize is available from the Adaption app, with access requiring authentication.

The announcement, authored by Simra