Alexandr Guzhva: Why I Joined Zilliz
My name is Alexandr Guzhva and I work in Zilliz on performance optimization.
I’ve joined Zilliz to outcompete the company’s competitors. And because I felt that I’ll be able to fully use my expertise.
I spent 15 years in finance, where I started using the approximate nearest neighbor search (ANNS) approach in order to speed up time series predictions for algo trading simulation, both for CPU and GPU. The computing engine that I’ve created could perform simulations several magnitudes faster than the baseline, thanks to ANNS.
After that, I spent about 2 years at Meta, where I’ve been adding performance-related code to the FAISS library and using its powerful facilities to apply vector codecs for compressing huge recommendation models that Meta uses for Ads. I became considered as one of the authors of the FAISS library because of my code donations.
Overall, I wrote more than 2M lines of code during my career and I’ve been dealing with similarity search methods for 10 years.
Zilliz as a company became interesting to me, because I knew that it uses various ways to speed up its products, such as an integration with NVIDIA Raft, experiments with various hardware platforms and advanced similarity search methods. This gave me the confidence that the required engineering and research activities are quite non-trivial, not python-monkey style ones :) A lot of companies are using it, which indicates that its products are competitive.
During the technical interview, the CTO of Zilliz, James Luan, told me that the company aims to be the top 1 in the vector database area, no other options are acceptable. Nowadays, vector databases is a very rapidly evolving field, and it implies a very high level of competitiveness. This particular phrase of James made me ultimately join Zilliz, because it matches my way of thinking and it is very familiar and natural to me. Also, I became curious, because the vector database field introduces additional challenges to the traditional similarity search applications, such as the necessity to modify the data or to include additional metadata with every sample.
I will continue donating my improvements to Milvus OSS and I plan getting more and more involved in making Zilliz products as best as possible. Maybe, I’ll be able to apply my knowledge of using ANNS for time series prediction at a certain moment of time.
Take Zilliz for a Spin for FreeGet Started Free
Share this article