Booking.com 与 Weaviate
2 分•作者: CShorten•大约 2 个月前
向量搜索看起来很简单，直到你达到生产规模。
我非常兴奋地分享 Weaviate 播客的新一期，与来自 @bookingcom 的 Başak 一起探讨生产规模的向量搜索、RAG 和基于 Agent 的 AI，由 @weaviate_io 呈现！
播客首先讨论了 Booking 采用向量搜索的临界点以及新兴用例。
仅 Partner-to-Guest 消息传递的规模就令人难以置信！ 每天就有近 25 万次此类交流，而 Booking 的 Agent 已经帮助处理了数万次！
Başak 描述了团队如何应对不断增长的规模和工作负载复杂性。他们对 Weaviate 进行了详尽的评估，使用了 1 亿个嵌入，并进行了通常被排除在常见 ANN 基准测试之外的测试。这包括过滤向量搜索、多线程并发，以及同时进行读写测试。
播客最后以 Başak 在 Booking 的职业生涯以及她对旅行社的看法作为结尾！
YouTube: https://www.youtube.com/watch?v=O9edM9ZS_FQ
Spotify: https://spotifycreators-web.app.link/e/8tc6Dyb7e3b
查看原文
Vector search looks easy, until you hit production scale.<p>I'm super excited to share a new episode of the Weaviate Podcast with Başak from @bookingcom on production-scale vector search, RAG, and agentic AI with @weaviate_io!<p>The podcast begins by discussing Booking's tipping point into adopting vector search and emerging use cases.<p>The scale of Partner-to-Guest messaging alone is insane! There are nearly 250,000 such exchanges <i>daily</i>, and Booking's Agent is already helping with 10s of thousands of these!<p>Başak describes how the team navigated increasing scale and workload complexity. They ran an exhaustive evaluation of Weaviate with 100M embeddings and tests often left out of common ANN benchmarks. This includes Filtered Vector Search, Multi-Threaded Concurrency, and testing with simultaneous Reads and Writes.<p>The podcast concludes with Başak's career journey to Booking and her thoughts on Travel Agents!<p>YouTube: https://www.youtube.com/watch?v=O9edM9ZS_FQ<p>Spotify: https://spotifycreators-web.app.link/e/8tc6Dyb7e3b