Skip to main content
AmigoFish

Mercado Libre turns data into a system of action to power commerce across Latin America

From Google Cloud · · 3 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.

If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.

Watch on YouTube

Not rated yet

Description

*Featured in this video:* Fernando Croceri, Director of Data & AI at Mercado Libre

*Executive summary:* With help from Google Cloud, Mercado Libre, Latin America's largest online commerce and financial services ecosystem, is building a free market for AI creativity. Using Google’s industry leading models — including Gemini, Veo, and Nano Banana — Mercado Libre developed a self-service AI media platform that helps employees create brand-aligned assets, build their own content pipelines, and run quality checks in real time. This new creative engine has helped teams achieve a 10x leap in productivity, producing 40 million images in just three months. Behind the scenes, the platform also connects to the company’s data ecosystem, built on BigQuery, Looker, and AlloyDB, opening new avenues to personalize experiences including virtual try-on features and a new tool that lets sellers automatically generate high-quality, contextualized product images.

*Challenge:* As Mercado Libre’s ecosystem grew, so did the demand for high-quality, brand-consistent media at speed. Fragmented tools and limited central visibility slowed production, made cross-team discovery harder, and led to overlapping work, highlighting an opportunity to unify creative workflows and use resources more efficiently.

*Solution:* Working with Google Cloud, Mercado Libre engineered an AI media platform that gives employees access to cutting-edge AI models like Gemini, Veo, and Nano Banana. Within the platform, employees can generate images and videos using built-in routing logic that automatically selects the optimal model for every task. To speed up content creation while maintaining strict brand standards, Mercado Libre offers pre-built templates for marketing campaigns, promotional banners, and product sliders. Underpinning this entire architecture is a data foundation built on Google’s Agentic Data Cloud — including BigQuery, Looker, and AlloyDB — that allows Mercado Libre to connect rich media with core enterprise data to speed product discovery, automate catalog matching, and power more personalized customer experiences.

*Results:* The new AI media platform is turning creative bottlenecks into competitive advantage. The impact is clear: a 10x surge in productivity and 40 million custom assets generated in just three months — paving the way for new interactive features like virtual try-on and content creation tools to empower entrepreneurs and small businesses across Latin America.

*Key takeaways and highlights from our interview with Fernando Croceri, Director of Data & AI at Mercado Libre:*

→ “Content creation was a major bottleneck — manual, fragmented, slow, with limited room for experimentation. The media platform that we developed allows us to generate content faster, at scale with less friction. You can edit images, create images, create a video, build pipelines, and we have some agents that check the consistency of every asset that is generated.”

→ “The biggest difference is not just speed, it's how our teams work. In the last three months, we created more than 40 million images, and at the moment, we are 10x productivity and we experiment 10x by using the media platform.”

→ “We have a gen AI gateway built by our team that allows us to connect to every cloud and every model. We, of course, use Google Cloud Gemini, Nano Banana, Veo. It’s not just the visual quality that they have, but it’s the control and the consistency that you have at scale.”

→ “Our users run more than 1,600 queries per second. So, it’s really a massive platform. It's not just storing data, it's about turning it into a system of action. We migrated to BigQuery. We also started using Looker as our semantic layer. You can do machine learning, call Gemini to process data, store embeddings, search for information.”

→ “AlloyDB is the key for handling real-time workloads. We store embeddings and serve in real time — less than 100 milliseconds at P95. It allows us to really search…

Back to Google Cloud · Report a problem