The pipeline

The pipeline is the sequence of stages a product goes through, from the moment it enters the system until it's published in your store. Each stage does one thing, has its own configurable autonomy level, and can pause or ask for approval.

The 6 stages

1. Normalization — a clean schema before touching anything
2. Research — optional, you turn it on in «Preferencias» (Preferences); it looks for sources on the web, reads the manufacturer's product page and saves each fact with its source
3. Content generation — description, SEO and attributes in your brand voice
4. Image processing — classification, adaptation, reconstruction, collages
5. Quality control (QA) — a 0-100 score and specific issues
6. Publishing — diff preview, sending and rollback to any version

Before writing, Noobit consolidates Product Knowledge: the facts from all your sources, each tagged with where it came from. The product page is written from there.

They run in order, but not all of them always run. If a product already has enough content, you can skip generation. If you're not publishing yet, the pipeline stops at QA and waits.

How to start it

There are three ways to start the pipeline for a product:

Autonomy levels per stage

Each stage is configured independently:

Automatic rules

Beyond the fixed autonomy level per stage, you can create rules that act on their own when something happens:

• “When a scan is linked, if it has scanner photos, rebuild the product photo”
• “When Product Knowledge is consolidated, if the product has no generated content yet, generate it”
• “When quality control finishes, if the score is below 70, notify the team”
• “When content is generated for a product with images, process its images”

To write in a different voice depending on the brand or collection, assign a voice to the group in «Base de conocimiento» (Knowledge base).

Images: from the supplier to your store

Noobit downloads the supplier's original photos in the background; if a supplier blocks any, Noobit flags them and you can retry them from «Importar CSV» (Import CSV). Photos that processing doesn't accept go to «Revisión imágenes» (Image review), where you accept or reprocess them. Size and watermark are applied when publishing, using what you set in «Preferencias»: the processed image doesn't change.

Traceability

Every pipeline run leaves a complete trail. In the product record, in the «Actividad» (Activity) tab (under the «Historial» (History) group), you see each operation and its steps, with the time, the result and the credits it used. Useful for audits and for spotting where the process slows down.

Failures and retries

If a stage fails (a provider that doesn't respond, a model that takes too long…), the pipeline stops and the stage is marked as failed. You can retry from the product record with «Reintentar pipeline» (Retry pipeline), without losing the earlier stages. Errors reach you as a notification in the bell in the header, with a link to the product.