Why it matters
Bogdan Gaza describes operational lessons from DatologyAI’s large synthetic-data pipeline: batch object-store metadata listing, size partitions to bound lost work, checkpoint partial outputs, and coordinate Ray CPU heads with GPU workers. Curated source documents seed rephrasing, while a shared workflow connects curation, generation, training and evaluation. The talk’s volume, throughput and model-quality figures are vendor-reported and lack a complete reproducible benchmark; the useful contribution is the breakdown of bottlenecks and recovery responsibilities.
My takeaway: Measure time spent discovering inputs before optimizing GPU inference. Simulate a late partition failure, check idempotent checkpoint writes and complete output coverage, and benchmark joint CPU/GPU scheduling under the actual batch workload.