10k+
Videos processed
Case Study | Rakuten GORA
Swing Analysis
Production backend delivery for a video-driven user feature, including ingestion, ML-team orchestration, media processing, release control, and multi-cloud resiliency.
Outcomes
~99%
Valid-input success
~30%
Processing-time reduction
100%
Release reliability
Problem
Swing Analysis launched as a net-new feature in the Rakuten GORA app. The backend platform needed to ingest and process user videos, coordinate requests to an external ML team pipeline, and return reliable results at production scale.
Architecture
Abstracted view of the Event-Driven Design and Multi-Cloud pipeline.
Ingestion Layer
Video Uploads & Metadata Storage (Azure MySQL)
→
CDC Engine
Debezium capturing Binlogs to Event Hubs
→
Orchestrator
ML Pipeline coordination & Media Processing (FFmpeg)
→
Delivery
Resilient Routing via Azure Front Door & AKS
EDA + CDC
Event-Driven Architecture using Debezium CDC to capture MySQL binlogs, ensuring reliable event propagation to the ML pipeline.
Storage + Eventing
Azure Blob Storage, Event Hubs topics, Event Grid antivirus events, and Azure MySQL.
Media Pipeline
C-based FFmpeg/JPEG processors wrapped by Java for video handling and frame extraction.
Deployment
AKS + Rakuten One Cloud, Azure Front Door routing, Argo Workflows, Argo CD, and Helm.
Ownership
- Owned backend integration across ingestion, orchestration, and result delivery paths.
- Engineered multi-cloud resiliency and routing behavior for production stability.
- Executed controlled release workflows with revision/image parameterized promotions.
- Created and modified Helm chart values for environment-specific deployment needs.