Case Study: Carousell achieves faster AI fraud detection and more reliable operations with Google Cloud

Carousell Creating an ecosystem of second-hand goods for a greener future with Google Cloud

Carousell, the online marketplace platform, needed to scale fraud detection across all user chats and improve the reliability of its growing infrastructure. Its previous custom-built model could only analyze 10–20% of conversations, was costly to scale, and took weeks to update, while older compute machines were causing performance spikes and consuming engineering time.

Looker helped by enabling Carousell to pair BigQuery and Vertex AI for bulk AI inference with Gemini models, allowing fraud analysis across the full dataset and reducing model iteration from three weeks to just a couple of days. Carousell also migrated workloads to C3D machines, which eliminated platform instability, cut related on-call burden to zero, and reduced compute cluster costs by 10%.

Carousell

Sanjeev Jaiswal

Senior Director of DevOps, SRE, Platform, and Cybersecurity Engineering