Skip to content
Digital media

Translation engine with a reviewer agent

Result5,000+articles translated in the first 24 hours
Context

A Brazilian outlet opened an international operation with its entire archive in Portuguese. The backlog had to be translated, and every new story after it, at the pace of a newsroom that doesn't stop to wait for a translator.

What we built

Two agents in a loop: one translates, the other grades the translation against the source and sends it back for correction until the score clears the bar. The language pair is a configuration parameter, so opening a new language doesn't mean touching the engine.

After delivery

In production ever since, translating each story as it publishes. Human review sits at the point where it decides: an editor watches the queue and steps in when a text calls for judgment.

Digital media

Per-reader news recommendation

ResultRetention and engagement upno figure published
Context

The homepage was the same for everyone. The client wanted each reader to find what interests them first, without the newsroom losing its say over what matters that day.

What we built

Human curation, automatic content tagging and a model that learns from reading behavior, combined into one queue. The desk pins what has to appear; the order of everything else is computed per reader.

After delivery

Both curves rose after the switch. The numbers belong to the client and aren't cleared for this page, so they stay in the conversation, under NDA.

Agribusiness fair

Custom artwork for the fair's partners

Result50+unique pieces under a single art direction
Context

The fair wanted to give each of its main partners a piece made for them. Every artwork had to be individual, all of them had to read as one visual family, and the deadline was the fair's own.

What we built

We locked the art direction first, palette, texture and framing, and only then generated the pieces against that reference, one at a time, with curation and adjustment at every round.

After delivery

Every partner received the piece made for them within the fair's deadline. The client credits the gift with a closer relationship to that group.

Next project

From your process to a result

We start with the process that needs to change and the measure that will show whether the delivery worked.

See how each line of work runs
From conversation to proposal

Proposal within 48 hours

After a 30-minute conversation, you receive the measure, scope, timeline, price, and limits in writing.

Conversation
30 minutes
Timeline
Within 48 hours
Format
3 pages maximum
First step

Describe the process that eats the most time today. We'll come back with two or three questions about it.

Discuss a project

Your data stays within the project. You decide whether any result can be published.