Charles Xie

Founder & CEO · Zilliz

Vector database and AI data infrastructure for machine learning and LLM applications15+ years experience

Charles Xie spent six years at Oracle as a founding engineer on the 12c database, where he concluded that the industry had solved structured data and almost entirely ignored the unstructured data that makes up the great majority of what organisations hold. He founded Zilliz to address that, and created Milvus, the open-source vector database released in 2019 that became core infrastructure for similarity search and, later, for retrieval-augmented generation in LLM applications. He sits on the board of the LF AI & Data Foundation and chaired it in 2020-2021.

Last reviewed: Sep 19, 2026

Charles Xie

Founder & CEO Zilliz

Last Reviewed: September 19, 2026

Executive Summary

Charles Xie’s career divides cleanly in two. For six years he was a founding engineer on Oracle’s 12c database, working at the centre of the relational database industry at the point where it moved to the cloud. The conclusion he drew from that work was that the industry had comprehensively solved a minority of the problem: structured data was well managed, while unstructured data — images, audio, video, documents, embeddings, on his account roughly ninety per cent of everything organisations hold — was barely analysed at all. He founded Zilliz to attack that, and created Milvus, the open-source vector database released in 2019. Milvus began as infrastructure for similarity search and became, with the rise of large language models, one of the standard components of retrieval-augmented generation — the mechanism by which an LLM is grounded in an organisation’s own data. He sits on the board of the LF AI & Data Foundation and chaired it during 2020-2021.

Career Highlights

  • Created Milvus, the open-source vector database, released in 2019.
  • Founded Zilliz, the company behind Milvus, around 2017-2018, naming it for the ambition of managing zillions of data.
  • Grew Milvus adoption from roughly 30 users at the end of 2019 to more than 200 during 2020, before the LLM wave made vector search a mainstream requirement.
  • Serves on the board of the LF AI & Data Foundation and chaired it during 2020-2021.
  • Spent six years at Oracle as a founding engineer on the 12c database, with more than fifteen years in databases overall.
  • Contributes writing on vector databases and AI data infrastructure to VentureBeat and gives regular interviews on the category.
  • Holds a master’s degree in computer science from the University of Wisconsin-Madison.

Professional Journey

Xie took a master’s in computer science at the University of Wisconsin-Madison — a department with a long lineage in database systems research — and joined Oracle, where he spent six years as a founding engineer on the 12c database. That release was Oracle’s move toward multitenancy and cloud delivery, which meant working on how a single database engine serves many isolated tenants efficiently. It is exacting systems work, and it gave him more than fifteen years of accumulated database experience by the time he left.

The insight he took away was about scope rather than technique. Decades of database engineering had gone into structured, tabular data, which by his estimate represents about a tenth of what organisations actually store. The rest — documents, images, audio, video, sensor output — sat outside the system of record, with only a small fraction ever analysed meaningfully. Bridging that gap required a different data structure entirely: not rows and indexes over exact values, but vectors and indexes over similarity.

He founded Zilliz around 2017-2018 to build it, choosing a name that gestured at managing zillions of data points. The company’s central artefact is Milvus, released as open source in 2019: a database purpose-built to store high-dimensional embeddings and retrieve nearest neighbours across billions of them at low latency. Open-sourcing it was the deciding strategic choice. A new category of database earns adoption by being used and trusted rather than sold, and Milvus’s growth reflected that — about thirty users by the end of 2019, more than two hundred through 2020, at a point when vector search was still a specialist requirement for recommendation and image-retrieval systems.

The category then changed underneath the company in its favour. The arrival of large language models made embedding-based retrieval a mainstream architectural need, because retrieval-augmented generation depends on exactly the operation Milvus was built to perform: finding the semantically nearest passages in a private corpus and supplying them to a model. Infrastructure built for similarity search in 2019 turned out to be the missing component of the 2023-onward AI application stack.

Alongside the company, Xie has worked on the governance side of open-source AI. He joined the board of the LF AI & Data Foundation, the Linux Foundation body that hosts and neutrally governs open AI and data projects, and chaired it during 2020-2021 — a position that matters in a field where enterprise adoption often depends on a project being seen as vendor-neutral rather than controlled by the company that wrote it.

Education

University of Wisconsin-Madison

Master of Science, Computer Science