Case Study: FireworkTV achieves 10× faster recommendations and 2× lower ML infrastructure costs with Provectus and Amazon SageMaker

FireworkTV rebuilds ML infrastructure 10x faster with Provectus

Firework (formerly FireworkTV), a video commerce and short video platform, faced challenges with its machine learning infrastructure as its platform scaled. Its initial setup, which ran on general-purpose compute with custom tooling, was slow and costly, hindering the speed and quality of its personalized video recommendations. To accelerate model training and serving while reducing costs, Firework engaged Provectus to migrate its ML workloads.

Provectus successfully migrated Firework's ML infrastructure to Amazon SageMaker in just four weeks. The solution provided distributed training pipelines and cost-optimized serving, which resulted in a 10x acceleration in model training and serving and a 2x reduction in ML infrastructure costs. This measurable performance gain led to a direct improvement in user engagement on the platform.

FireworkTV