Speedtrain is for repeatable content jobs over many records, not one-off writing. If you need one product description, write it. If you need 40,000 of them in four languages, each one checked against your brand rules, that is a task configuration.
What you do with it
Enrich records
Turn sparse product data into descriptions, bullet points, and marketing copy.
Classify and structure
Assign categories, extract attributes, and normalise messy values.
Translate and localise
Produce the same content per market, with per-language rules.
Read images
Let a vision model describe what is actually in the product photo.
The governance model
Speedtrain assumes AI output needs checking. Three mechanisms do that work, and they run on every single item:1
Quality checks score the output
Deterministic rules (schema validation, forbidden terms) run as code in milliseconds. AI rules handle the questions that genuinely need judgment. Each produces a score from 1 to 5 and written feedback. Quality checks
2
A quality gate aggregates the scores
The gate is arithmetic, not an opinion — no model is involved. It decides whether an item can auto-approve or must go to a person. Quality gates
3
People decide
Every item lands in a review session where a human can accept, decline, or send it back for another attempt. Nothing reaches a destination without passing through this. Review sessions
Where to go next
Data flow
How data moves between Enterspeed and Speedtrain, in four steps.
How it works
The full path from source data to published content.
Getting started
Build and run your first task configuration.
Key concepts
The vocabulary, one page per concept.