Building ML Platform Capabilities at Global Scale
summary
Prosus teams build distinct platform pieces that actually run at scale: OLX pairs a searchable data lake and Airflow-driven reservoirs with KFServing for real-time inference to serve image and attribute models; iFood wraps AWS SageMaker with a shared library, runs a Databricks + Spark feature store served from Redis and automates daily model CI/CD; Swiggy’s homegrown JVM platform handles peaks over 300k qps and produces over a billion predictions a day while keeping p95 latencies around 20–25 ms.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:21Global Consumer Internet Group
- 5:17Olx
- 7:57Machine Learning Workflow
- 9:34Scheduler
- 17:04Swiggy
- 19:13Architecture
- 21:30General Considerations
- 22:17Use Multiple Things in Parallel
related talks
30mML Platform for Insurance Conglomerate
26mManaging ML Models @ Scale
28mBuilding Developer-Centric ML Inference Platforms
23mML Platform Challenges & Applications at Freshworks
31mUse-Case Driven ML Platforms
48mFrom (Big) Data Mess to Data as an Innovation Enabler
27mML Platform Journey
13mBuilding Scalable AI/ML Platforms for Industrial IoT