5.1. Possibilities offered by enabling LLMs

5.1.1. Line-by-line AI: leveraging AI at an industrial scale

Once an LLM is connected, you can systematically invoke the AI on every row of the tables you wish to process using the AI Gen node. Use cases are virtually limitless:

Sentiment analysis
Text generation or reformulation
Structured information extraction
Intelligent classification
Application of business rules requiring context or common sense
Normalization and semantic enrichment

All the capabilities offered by language models can thus be applied at an industrial scale to your data, directly within your processing flows. This usage consumes tokens for processing each row, unlike usage in the preparation editor below.

5.1.2. Natural language transformations: frugal AI, predictable processing without token consumption

In the preparation editor, you can develop natural language transformations. The LLM then intervenes only when developing the transformation.

Once defined and validated:

the LLM is no longer called upon during data processing
the transformation executes in a deterministic manner
performance remains high
no tokens are consumed in production.

You thus benefit from the comfort and flexibility of generative AI during design, while retaining speed, predictability, and cost control during execution.

5.1.3. AI-enriched flash audit

Discover how to instantly evaluate the health of your data with Tale of Data. By simply clicking the Quality audit button, the generative AI analyzes a targeted sample of your dataset (limited to around a hundred rows) to draw up an exhaustive list of detected anomalies before any modification.

For maximum efficiency, first validate your transformations on this limited sample — which you can refresh at any time using the Refresh sample button — before automatically applying them to your entire dataset during flow execution.

Warning

Depending on the bandwidth of your generative AI service, report generation may take from a few seconds to a little over a minute.

5.1.4. Documentation automatique des flows

Your LLM can also assist you in documenting your flows. It can generate, complete, or rephrase descriptions, in the language of your choice.

This facilitates collaboration, improves maintainability, and guarantees clear documentation, even in international environments.

Note

To learn more, you can consult the tab Afficher et masquer la documentation sur le canevas at any time.

5.2. Data security and control

5.2.1. Local or remote LLM

The integration of an LLM into Tale of Data is aligned with a logic of control and compliance. You retain complete control over your technology choices and data flows.

You can connect the platform to the AI of your choice, provided it is approved by your company. Tale of Data does not impose any provider on you: you remain free to use the model that matches your security, performance, and regulatory compliance requirements.

Several deployment options are possible:

Using a cloud-based LLM
Using an on-premise deployed LLM
Provision of a dedicated LLM by Tale of Data (contact the sales team for more information)

This flexibility allows you to align the use of generative AI with your internal data governance policies.

5.2.2. Control over the scope of shared data

The use of an LLM comes with clear transparency regarding the scope of transmitted data. The volume and nature of the sent information depend on the activated use case:

Line-by-line AI: all data involved in the processing is transmitted to your AI to enable the requested analysis or enrichment.
Flash audit and preparation editor: a sample of rows is sent to your AI, strictly limited to the needs of analysis or design assistance.
Flow documentation: only the flow (structure, steps, metadata) is shared with your AI, without transmitting underlying business data.

You thus have a clear view of what is shared, when, and for what purpose. Tale of Data allows you to leverage the power of generative AI while maintaining rigorous control over your data.