MODO

Where they were

MODO is the digital wallet backed by Argentina’s leading banks: it has more than 4 million users and 350 employees, 20 of them on the data team. When they reached out, the goal was to accelerate time to market for models in production. The work grew from there into architecture, data availability, and GenAI features.

What we found

The real problem wasn’t just technical. They weren’t missing an ML engineer; they were missing a complete operational framework to guide model training and deployments. From there we went deeper into architecture and data availability.

What we built

Three contracts, three different problems, the same team. In the first, we designed and implemented a complete MLOps framework: training pipelines, automated deployment, monitoring, and retraining. In the second, we contributed to the development of Contextual Search, a generative AI feature that reached production a year before the competition, with zero incidents since launch. In the third, we worked on reducing latency and cost of LLMs in production, achieving savings of thousands of dollars per month in infrastructure through code refactoring and architecture updates.

What they can do now

MODO operates the MLOps framework independently. The internal team trains, deploys, and monitors their own models without depending on deployr, and they have a development environment we set up so they can experiment without risk.

Infrastructure savings are measured in thousands of dollars per month. Generative AI features serve more than 4 million users in production.

Why they came back

Three separate contracts. Every time MODO had a new technical challenge they couldn’t solve internally, they trusted us to help them. That’s what we aim for: not dependency, but trust.