Case Study: Mercari scales GenAI success with Weights & Biases

Mercari scales GenAI with Weights & Biases, reaching 22,000 runs in 2 weeks

Mercari, Japan's largest C2C marketplace, sought to integrate generative AI applications to improve its buyer and seller experience and internal services. Their challenge was to iterate quickly on these new GenAI apps, validate their outputs, and gain organizational buy-in without a clear initial strategy for fine-tuning their own models. They partnered with Weights & Biases, leveraging its W&B Weave platform for evaluation and observability.

Using Weights & Biases' tools, Mercari adopted an evaluation-centric development approach. This enabled rapid iteration, prompt and model comparison, and robust versioning and reproducibility. The solution allowed their team to quickly validate product-market fit, fine-tune a more effective and cost-efficient model, and successfully deploy multiple GenAI applications. The result was a significant acceleration in their development cycle, with one engineer achieving over 22,000 tracked runs in two weeks, leading to scalable success and widespread organizational adoption of GenAI.

Mercari