Skip to main content
Speedtrain is a governed platform for agentic data enrichment. You point it at structured business data — products, articles, records — describe the content you want generated, and it produces that content for every item, scores each output against your own quality rules, and routes the results to people for approval. It is built on Enterspeed, so the data you enrich and the content you produce move through the same data supply chain as the rest of your stack.
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
Humans are on the loop, not in it. You are not editing 40,000 descriptions — you are inspecting the ones the gate flagged, and spot-checking the ones it passed.

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.