Case Study: Ikigai Labs achieves scalable, interactive AI workflows with Anyscale Ray Serve

How Ray and Anyscale make it easy to do massive-scale machine learning on aerial imagery

Dendra needed a way to power highly interactive, spreadsheet-like AI workflows that could handle arbitrary Python code while still scaling to large datasets. Their challenge was to keep data pipelines mission-critical, transparent, and instantly browsable without sacrificing performance or creating painful dependency conflicts.

Anyscale helped Dendra build this on Ray Core and Ray Serve, enabling distributed execution, custom environment management, low-latency HTTP endpoints for “peek” operations, version-aware deployments to avoid race conditions, and traffic-based scaling of popular services. The result was a scalable, interactive data pipeline with sub-second browsing of intermediate data and smoother collaboration for users, with the case study noting a ~10-second task submission overhead was eliminated for peek runs.

Dendra

Richard Decal

Lead ML Engineer