It seems that vector databases are at the peak of their popularity, as AI continues to dominate the technology landscape. The rise of Large Language Models (LLMs) and the Generative AI (GenAI) movement has created a fertile environment for vector database technologies to thrive.
“Without using vector similarity search, you can still develop AI/ML applications, but you would need to do more retraining and fine-tuning,” Andre Zayarni, CEO and co-founder of vector search startup Qdrant, explained to TechCrunch.
In the world of data storage, traditional relational databases like Postgres or MySQL reign supreme for structured data. However, when it comes to handling unstructured data such as images, videos, emails, and social media posts, these databases fall short. This is where vector databases come in, as they store and process data in the form of vector embeddings.
These vector embeddings are numerical representations of data that capture the relationships between different data points. This makes them ideal for machine learning, as they can store data spatially based on relevance, allowing for easier retrieval of semantically similar data.
This technology is especially useful for LLMs, like OpenAI’s GPT-4, as it helps the AI better understand the context of a conversation by analyzing previous similar conversations. It also has practical applications for real-time tasks, such as content recommendations on social media and e-commerce platforms, as it can quickly retrieve similar items based on a user’s search query.
Aside from aiding in AI development, vector search can also prevent “hallucinations” in LLMs by providing additional information that may not have been available in the original training dataset.
“Dedicated [vector] databases tend to be fully focused on specific use cases and hence can design their architecture for performance on the tasks needed, as well as user experience,” explained Peter Zaitsev, founder of Percona, a database support and services company.
While vector databases have seen significant growth in recent years, with startups like Qdrant, Vespa, Weaviate, Pinecone, and Chroma securing millions in funding, larger database companies are starting to take note as well.
Earlier this year, Index Ventures led a $9.5 million seed round into Superlinked, a platform that converts complex data into vector embeddings. And just a few weeks ago, Y Combinator’s Winter ’24 cohort was announced, which included Lantern, a startup that offers a hosted vector search engine for Postgres.
One of the most noteworthy vector database startups, Marqo, raised $4.4 million and $12.5 million in seed and Series A funding respectively, with plans to provide a full range of vector tools via a single API.
“Working with visual search and robotics at Amazon was when I really looked at vector search — I was thinking about new ways to do product discovery, and that very quickly converged on vector search,” stated Jesse N. Clark, co-founder of Marqo.
However, while vector databases may be experiencing a moment in the midst of the AI hype cycle, they are not a one-size-fits-all solution for enterprise search scenarios.
“Users who are building very complicated and large-scale AI applications will use dedicated vector search databases, while folks who need to build a bit of AI functionality for their existing application are more likely to use vector search functionality in the databases they already use,” predicted Zaitsev.
But Qdrant and other vector database startups are confident that their native solutions, built solely around vectors, will provide the necessary speed, memory safety, and scalability as vector data continues to grow.
Their pitch is, ‘we can also do vector search, if needed,’” clarified Zayarni. “Our pitch is, ‘we do advanced vector search in the best way possible.’ It is all about specialization. We actually recommend starting with whatever database you already have in your tech stack. At some point, users will face limitations if vector search is a critical component of your solution.”
