Charles Xie – Founder & CEO of Zilliz and Creator of Milvus

Who Is Charles Xie?

Charles Xie is the founder and CEO of Zilliz and the creator of Milvus, an open-source vector database designed for large-scale similarity search. His career sits at the intersection of database engineering, distributed systems, cloud infrastructure, and artificial intelligence.

Before founding Zilliz, Xie spent approximately six years at Oracle's US headquarters, where he worked as a founding engineer on the Oracle 12c cloud database project. His experience building large-scale database infrastructure helped shape his understanding of distributed systems and cloud-native data management.

In 2017, Xie founded Zilliz with the goal of building new infrastructure for unstructured data. The company developed Milvus, a database designed to store, index, and search vector embeddings at scale.

Education

Charles Xie's academic background is in computer science. He earned a Master's Degree in Computer Science from the University of Wisconsin-Madison and a Bachelor's Degree from Huazhong University of Science and Technology.

According to the available source material, Xie also began a Ph.D. program at the University of Wisconsin-Madison but left the program before completion to pursue his professional career.

Early Career at Oracle

Before becoming an entrepreneur, Xie built his expertise in database systems at Oracle. He spent roughly six years at Oracle's US headquarters and was involved as a founding engineer in the Oracle 12c cloud database project.

Oracle 12c introduced important cloud-oriented database capabilities, including its multitenant architecture. Working on a large enterprise database project gave Xie experience with database architecture, distributed systems, cloud infrastructure, scalability, and enterprise data management.

This experience later became important when he began thinking about a different category of data: unstructured data.

The Unstructured Data Problem

Traditional databases are highly effective at managing structured information such as rows, columns, transactions, numbers, and categorical values. However, images, videos, audio recordings and large collections of text do not naturally fit the traditional relational database model.

The emergence of artificial intelligence changed this situation. AI models could transform unstructured information into numerical representations called embeddings. These vectors could represent the semantic characteristics of text, images, audio and other forms of data.

Xie recognized that these embeddings represented a new type of data that required specialized infrastructure for storage, indexing and retrieval.

Founding Zilliz

Charles Xie founded Zilliz in 2017. The company initially operated from Shanghai and later expanded its global presence, with San Francisco becoming its headquarters.

The company's central idea was to build infrastructure capable of managing massive quantities of vectorized data. Rather than treating AI-generated vectors as an unusual data format, Zilliz approached them as a fundamental database data type.

The initial challenge was significant because there was no established blueprint for building a production-grade vector database at billion-scale.

The Birth of Milvus

Milvus emerged from Zilliz's effort to build a specialized database for vector data. Development began around 2017, with early versions focused on vector search and the management of high-dimensional data.

Early users included engineers and researchers working on machine learning applications who needed infrastructure for managing embeddings and similarity search.

Milvus was eventually released as open source in November 2019 under the Apache 2.0 license. At the end of 2019, the project had approximately 30 users. That number grew to more than 200 in 2020 and approached 1,000 soon afterward.

Why Milvus Was Different

Milvus was designed as a database rather than simply a similarity-search library.

1. It provides native support for vector data.
2. It supports specialized approximate nearest-neighbor indexes.
3. It provides similarity search using distance metrics.
4. It supports distributed deployments and horizontal scaling.
5. It supports metadata filtering and hybrid search.
6. It is designed to operate with very large vector collections.

This distinction separated Milvus from libraries such as FAISS, which provide highly optimized similarity-search algorithms but do not provide the complete database infrastructure required for persistence, distributed operations, concurrent workloads and production deployment.

Open-Sourcing Milvus

In November 2019, Zilliz open-sourced Milvus under the Apache 2.0 license.

The open-source strategy lowered the barrier for developers to experiment with vector search while creating a mechanism for users and contributors to provide feedback, integrations, documentation and code.

The strategy also established two complementary directions for Zilliz: Milvus as an open-source database and Zilliz Cloud as a managed commercial service.

LF AI & Data Foundation

Milvus became part of the LF AI & Data Foundation, a Linux Foundation organization supporting open-source AI and data technologies.

Charles Xie served as chairperson of LF AI & Data in 2020 and 2021 and continues to serve on its board. Milvus subsequently graduated from the foundation, reflecting the project's growing maturity and community.

Milvus 1.x and Its Limitations

The first-generation Milvus architecture used a single-master, multiple-slave design. It provided read and write nodes, shared storage and segment-based data organization.

Although this architecture supported early use cases, the increasing scale of vector workloads exposed several limitations.

1. Write bottlenecks caused by a single writer per collection.
2. High storage costs associated with NAS-based persistence.
3. Limited built-in high availability.
4. Operational complexity for distributed deployments.
5. Scaling and rebalancing challenges.

The Zilliz team eventually concluded that these problems could not be adequately solved through incremental changes. A fundamentally different architecture was required.

The Milvus 2.0 Rebuild

One of the defining moments in the history of Milvus was the decision to rebuild the system from the ground up.

Milvus 2.0 introduced a cloud-native distributed architecture based around several principles, including separation of computing and storage, microservices, log-based data processing and unified batch and stream processing.

Cloud-Native Architecture

Milvus 2.0 separates compute and storage so that different parts of the system can scale independently. It also separates workloads such as querying, data persistence and index construction.

Logs as Data

The log broker became a central component of the architecture. Data operations flow through the log system while worker nodes consume and process those logs.

Distributed Components

The architecture includes access services, coordinator services, worker nodes and storage infrastructure.

The coordinator layer manages different aspects of the system, while Data Nodes, Query Nodes and Index Nodes perform specialized workloads. Metadata can be stored using etcd, logs can be handled through systems such as Pulsar, and object storage such as MinIO or Amazon S3 can store persistent data and indexes.

Vector Indexing

Vector indexing is central to Milvus's performance. Different indexes provide different balances between search speed, recall, memory consumption and storage requirements.

1. FLAT: Exact search with high computational cost on large datasets.
2. IVF: Partitions vectors and searches selected regions.
3. HNSW: Uses a graph structure for fast approximate nearest-neighbor search.
4. Product Quantization: Compresses vectors to reduce memory requirements.
5. DiskANN: Enables large-scale vector search using disk-based infrastructure.
6. GPU indexes: Accelerate suitable vector workloads using GPUs.

Hybrid Search

Milvus supports hybrid search by combining dense and sparse representations.

Dense vectors are useful for understanding semantic meaning, while sparse representations are effective for matching specific terms and keywords. Combining the two approaches can produce more relevant retrieval results than relying exclusively on either semantic or lexical search.

Multimodal Search

As AI systems increasingly work with text, images, audio and video, vector databases have become important infrastructure for multimodal retrieval.

When different types of content are represented as vectors, similarity search can be used to connect information across modalities. For example, a text query can be used to find visually similar images when the underlying models produce compatible representations.

Zilliz Cloud

Zilliz Cloud is the managed commercial offering built around Milvus. It provides managed infrastructure so organizations can use vector search without operating the underlying distributed database themselves.

While Milvus provides an open-source, self-managed option under Apache 2.0, Zilliz Cloud provides managed infrastructure, scaling, operational support and enterprise-oriented capabilities.

Business Model

Zilliz follows an open-core model. Milvus remains an open-source project that drives adoption and community development, while Zilliz Cloud provides a commercial managed service.

This approach allows developers and organizations to deploy Milvus independently while giving enterprises an option to use managed vector infrastructure.

Funding and Company Growth

According to the supplied research, Zilliz raised approximately $115 million across seven funding rounds.

1. 2017: $2.5 million seed funding
2. 2018: $10 million Series A
3. 2020: $43 million Series B
4. 2022: $60 million Series B extension

The company expanded from its original Shanghai base to a global organization headquartered in San Francisco.

The Generative AI Revolution

The emergence of large language models dramatically increased the importance of vector databases.

Before the generative AI boom, vector search was primarily associated with recommendation systems, image search and specialized machine learning applications. The rise of large language models made retrieval infrastructure central to a much wider range of AI applications.

RAG systems require a retrieval layer capable of finding relevant information from large collections of documents. Vector databases such as Milvus became an important part of this architecture.

Milvus and RAG

Retrieval-Augmented Generation combines information retrieval with large language models.

1. A user submits a question.
2. The question is converted into an embedding.
3. Milvus searches for similar vectors.
4. Relevant documents are retrieved.
5. Retrieved information can optionally be reranked.
6. The language model generates an answer using the retrieved context.

This approach allows organizations to connect AI models with private, specialized or frequently updated information without retraining the underlying model every time the knowledge base changes.

Milvus and AI Agents

AI agents increasingly require persistent memory and retrieval capabilities. Vector databases can provide a storage and retrieval layer for information that agents need to access across tasks and sessions.

Milvus can therefore serve as part of an agent infrastructure stack, helping applications retrieve relevant memories, documents and knowledge during an agent's reasoning process.

What Did Charles Xie Actually Invent?

Charles Xie's role should be understood as a founder, technical visionary and architectural leader rather than as the sole person who individually coded every component of Milvus.

His documented contributions include recognizing the need for specialized infrastructure for vectorized unstructured data, founding Zilliz, defining the vision behind Milvus, leading the engineering organization and guiding major architectural decisions.

The Milvus database itself was built through the work of the broader Zilliz engineering team, together with contributions from the open-source community.

This distinction is important. Milvus is a team-built open-source system, while Charles Xie provided the entrepreneurial leadership, product vision and technical direction that helped turn the concept into a major database project.

Milvus's Impact on AI Infrastructure

Milvus contributed to the establishment of vector databases as an important category of AI infrastructure.

Its open-source development helped make large-scale vector search accessible to developers and organizations. As RAG, semantic search, recommendations, multimodal retrieval and AI agents expanded, the role of vector databases became increasingly important.

Milvus demonstrated that vector search could be implemented as a distributed database rather than simply as an isolated machine-learning library.

Current Milvus

According to the supplied research, Milvus has evolved into a mature distributed vector database supporting large-scale vector workloads, multiple indexing approaches, hybrid search, multimodal applications and cloud-native deployment.

The project remains open source under the Apache 2.0 license and has developed a large developer community.

Current Zilliz

Zilliz operates around two major offerings: Milvus, its open-source vector database, and Zilliz Cloud, its managed commercial vector database service.

The company's broader vision is to build infrastructure capable of managing and making useful the enormous quantities of unstructured data generated by modern organizations and AI systems.

Charles Xie Today

Charles Xie continues to serve as CEO of Zilliz and remains involved in the broader AI and open-source infrastructure ecosystem. His work focuses on the intersection of databases, vector search, AI infrastructure and unstructured data.

He also continues to contribute to discussions around the evolution of vector databases, retrieval systems, multimodal AI and the future of data infrastructure.

Complete Timeline

Interesting Facts About Charles Xie and Milvus

1. Charles Xie founded Zilliz in 2017.
2. He previously worked at Oracle's US headquarters for approximately six years.
3. He was a founding engineer on the Oracle 12c cloud database project.
4. Milvus was open-sourced in November 2019.
5. Milvus uses the Apache 2.0 open-source license.
6. Milvus evolved from a standalone vector search system into a distributed cloud-native database.
7. Milvus 2.0 introduced separation of compute and storage.
8. Milvus supports both dense and sparse vector search.
9. Milvus is used in RAG and other AI retrieval architectures.
10. Zilliz operates both Milvus and the managed Zilliz Cloud service.
11. Charles Xie served as chairperson of LF AI & Data in 2020 and 2021.
12. Milvus has become one of the major open-source projects in the vector database ecosystem.

Frequently Asked Questions

Who is Charles Xie?

Charles Xie is the founder and CEO of Zilliz and the creator of Milvus, an open-source vector database designed for large-scale similarity search.

What is Milvus?

Milvus is an open-source vector database designed to store, index and search high-dimensional vector embeddings at scale.

Who built Milvus?

Milvus was built by the Zilliz engineering team under the leadership of Charles Xie, with contributions from the open-source community.

Why was Milvus created?

Milvus was created to provide specialized infrastructure for managing and searching vector embeddings generated from unstructured data such as text, images, audio and video.

What is the difference between Milvus and FAISS?

FAISS is a similarity-search library, while Milvus is a complete database system designed for persistence, distributed operation, concurrent workloads and production-scale vector search.

Is Milvus open source?

Yes. Milvus is open source and distributed under the Apache 2.0 license.

What is Zilliz Cloud?

Zilliz Cloud is the managed commercial offering built around Milvus, allowing organizations to use vector database infrastructure without managing the underlying system themselves.

What is RAG?

Retrieval-Augmented Generation combines information retrieval with language-model generation. Vector databases such as Milvus can provide the retrieval layer for RAG systems.

Why is Milvus important for AI?

Milvus provides scalable vector retrieval infrastructure for applications including semantic search, RAG, recommendations, multimodal retrieval and AI agents.

What did Charles Xie do at Oracle?

Charles Xie was a founding engineer on the Oracle 12c cloud database project and developed substantial experience in database architecture and distributed systems.

Where is Zilliz headquartered?

Zilliz is headquartered in San Francisco, according to the supplied research.

Conclusion

Charles Xie's career illustrates the convergence of database engineering and artificial intelligence. After years of working on enterprise database infrastructure at Oracle, he identified a growing problem: traditional databases were not designed for the enormous volumes of unstructured information being transformed into vector embeddings by AI.

In 2017, he founded Zilliz and began building Milvus around this idea. The project evolved from an early standalone vector search system into a distributed, cloud-native database capable of supporting large-scale vector retrieval.

Open-sourcing Milvus in 2019 helped accelerate its adoption and created a broader community around the project. The later rebuild into Milvus 2.0 demonstrated the importance of rethinking architecture when workloads and scale change fundamentally.

The rise of generative AI and RAG ultimately turned vector databases from a specialized technology into an important component of modern AI infrastructure. Milvus is now used as a retrieval layer for applications ranging from semantic search and enterprise knowledge systems to multimodal AI and agent memory.

Charles Xie's story is therefore not simply the story of one inventor creating one product. It is the story of a technical vision translated into an open-source project by a larger engineering team and community. Xie provided the vision, leadership and architectural direction, while the Zilliz team and open-source contributors built Milvus into the infrastructure platform it is today.