iFood, the Latin American food-tech leader processing millions of orders per month from São Paulo, Brazil, built a centralized ML platform on Amazon SageMaker that now powers hundreds of production models for personalized recommendations, real-time delivery tracking, automated customer service, and grocery shopping assistance. "At iFood, we were at the forefront of adopting transformative machine learning and AI technologies, and our partnership with the SageMaker Inference product team has been instrumental in shaping the future of AI applications," Daniel Vieira, ML Platform manager at iFood, said in a joint AWS blog post.
For years iFood's data scientists trained models in notebooks and threw them over the wall to engineering teams, who struggled to integrate them into production — a hand-off pattern that slowed time-to-deployment for customer-facing features. The new internal platform, called ML Go!, bridges data science and engineering by standardizing training and deployment through Amazon SageMaker Pipelines, SageMaker Inference Components, and BruceML, an internal scaffolding tool that wraps model code into Docker containers ready for SageMaker. Every trained model is automatically registered in the SageMaker Model Registry with versioning; real-time predictions run on hosted endpoints, while offline predictions use SageMaker batch transform.
The platform powers four customer-facing experiences. Personalized restaurant and menu recommendations analyze past orders and contextual signals to suggest dishes tailored to each customer's taste. Intelligent order tracking uses traffic, restaurant prep time, and courier location to predict delivery time with high accuracy. An AI chatbot handles thousands of daily customer inquiries in natural language. A grocery-shopping assistant accepts voice or text recipes and returns a curated shopping list. Beyond customer features, iFood's team co-developed with AWS several production enhancements — Inference Components, multi-model endpoints on GPU, asynchronous inference, and scale-to-zero — reducing foundation-model deployment costs by 50% on average and latency by 20% on average for SageMaker customers broadly.
iFood has also built ML Go! Gateway, a single abstraction for connecting applications to SageMaker endpoints and jobs, and uses Inference Components to separate concerns within endpoints so that models are served faster and more efficiently.