Frequently Asked Questions

PrettyWhale.ai is an Engineering AI for ingestion and integration code. You give it one sample of a data source and it generates a complete production-ready data pipeline: executable code, unit tests, data quality checks, schemas, configuration and documentation. In minutes rather than sprints.

We aim to build the future of AI for Data engineering.

PrettyWhale.ai is accessible through our Visual Studio Code extension.

PrettyWhale.ai is really different from copilots. Those tools work inside your editor and answer the question you are typing. They are good at it but they need you as a guide. The unit they produce is a suggestion, and an engineer decides what to do with it.

PrettyWhale.ai produces the unit above that one: a project you can run, with the tests that prove it runs and the documentation that lets someone else maintain it. Copilots suggest. Engineering AI delivers.

We wrote the full comparison, including where copilots win, in Engineering AI, Copilots and Code Generators: A Taxonomy.

No, and the distinction matters.

An ETL platform gives you a configured job, often running on a runtime you do not operate. As for us, we give you a complete data pipeline project in your repository, with its tests, its output schema and its documentation. And that code will run on the infrastructure of your choice.

The rest follows from that. Your data never passes through us. We sit outside your production path. The pipeline keeps running on your infrastructure the day you stop being a customer, and it goes where some vendor configuration cannot: on-premise, another cloud, a different orchestrator.

The sources differ too. A catalogue covers what it covers. The work we generate also covers everything else: an in-house ERP, a partner file, an export nobody documented.

Only a representative sample of the data source. Not your whole warehouse.

The sample needs to contain all the fields you actually deal with, including the ugly ones: the booleans stored as “yes” and “No”, the dates written out in plain text, the coordinates that arrive as a single string, and so on.

Field by field. Once your sample is analysed, each field gets its own transformations (normalization, conversion, validation, etc.). The model proposes the transformations needed to ensure the quality of output data. You can then review them, field by field, at the Processings stage.

Yes. At the end of the generation cycle, the pipeline is executed on our infrastructure before it is packaged. It runs on the sample you provided and the tests are also executed. Thus, what lands in your repository has already been executed.

That step is what separates our generated pipelines from a copilot’s suggestion. A suggestion is first judged on whether it looks right. Our pipelines are judged on whether they run, and we would rather find that out before you do.

A small language model (SLM) we trained ourselves on ingestion and integration code. No general-purpose model from another vendor is called at any point in the generation. That has a direct consequence for you: your sample is read by our model and by nothing else, and no third party sets the price, the terms or the behaviour of the thing that writes your production code.

The solution generates all the following artifacts, in one bundle with the same shape every time:

• Executable code
• Schemas and metadata parameters
• Unit tests
• Configuration files
• Documentation

PrettyWhale.ai is a French company, registered in France and governed by French law. The platform runs on infrastructure hosted in France. Your sample is processed by our own model, and since no third-party model is called, nothing leaves that perimeter.

Hosting details and data processing terms are available for your security and procurement teams on request.

No. Your sample is used to generate your project and nothing else. It does not become training material, for us or for anyone else.

No. The generation runs from a sample file. The generated code then connects to your systems when you deploy it on your infrastructure, with credentials you provide.

If your sample contains personal data, you can anonymize it before uploading, because the generation needs the shape of the fields rather than their real content.