Case Study: WeRide achieves 90x ML productivity gains with LanceDB

WeRide boosts ML productivity 90x with LanceDB

WeRide, a leading autonomous driving company, faced the challenge of efficiently mining massive amounts of complex, multimodal data from its fleet to find rare scenarios for model training. Their data infrastructure team needed a solution that could deliver millisecond search latency, support hybrid queries, and minimize I/O for their high-speed storage system. They partnered with vendor LanceDB to use its vector database to power the search engine within their revolutionized data platform.

LanceDB provided the vector search capabilities that enabled WeRide to quickly search for data using text-to-image or image-to-image queries across billions of data points. The implementation resulted in a 90x improvement in ML developer productivity for data exploration and a 3x reduction in ML training time by improving data I/O. Most significantly, the time required for data mining safety-critical edge cases was reduced from one week to just one hour. LanceDB's solution proved to be cost-effective, easy to maintain, and scalable for WeRide's growing data needs.

WeRide