Use cases · Catalogue content

How it works

Suppliers send product information as free-text paragraphs, inconsistent spreadsheets and PDF datasheets. The model reads each one and fills a fixed field list you define — material, dimensions, colour, EAN, care instructions, country of origin — and returns structured rows. The fast small model handles the bulk of the work; rows it flags as uncertain get a second pass from the large model or from a person.

Data you need

The supplier text itself: emails, description fields, PDF datasheets converted to text. Scanned or image-only PDFs are out of scope — no vision model is served — so those need OCR on your side first, or ask the supplier for the digital original. You also need a written definition of your target fields with allowed values. If the information genuinely is not in the supplier text, no model can extract it, so audit a sample before committing.

What to expect

Strong on the tedious bulk of extraction from readable text, and honest about uncertainty when set up to flag it. The usual failure point is not the model: complex table layouts in PDFs often mangle during text extraction before the model ever sees them. Numeric unit conversions (inches to centimetres) should be done in code, not left to the language model. Expect per-supplier tuning — each supplier's format fails in its own way, so accuracy settles per source, not globally.

Where people stay involved

A person spot-checks a random sample of every supplier batch — not only the rows the model flags, because the model's own confidence misses errors it is sure about. Fields that feed filters and legal labels (fibre composition, age warnings) deserve full review until you trust the pipeline on that supplier.

• Turn supplier spec sheets into first-draft product descriptions

  Drafts product descriptions overnight from the supplier data you already hold, in your shop's voice — a review job instead of a from-scratch writing job.

• Sort new products into the right category in your shop

  Embeddings shortlist the most plausible categories for each new product; a small model picks one and explains why — workable even on large category trees.

• Write page titles and meta descriptions for every product and category page

  Drafts an SEO title and meta description for every page from real product attributes, within your character limits — economical even at long-tail scale.

• Draft alt text for product images from the product data you already have

  Composes alt text from product titles, attributes and image metadata — a bulk improvement over empty alt attributes, not true image description.