Pinecone Systems Inc. is emerging from stealth mode today armed with $10 million in seed funding and a serverless vector database that it says can make machine learning queries much faster and more accurate.

The investment was led by Wing Venture Capital. Its founding partner Peter Wagner, who previously invested in cloud data warehouse company Snowflake Inc., will take a seat on Pinecone’s board of directors.

Pinecone was founded by its Chief Executive Edo Liberty, who previously ran Yahoo Inc.’s Scalable Machine Learning Platforms group and later led a team that helped build the Amazon SageMaker machine learning service at Amazon Web Services Inc.

The company argues that a vector database architecture is more suitable for machine learning due to some basic fundamentals. Machine learning models take data such as documents, videos or user behaviors, and convert this into vectors, which are complex collections of numbers. Performing inference is often a question of finding which vectors are nearest or most similar to others.

Pinecone says that in order to sort through and rank large numbers of vectors, a specialized data infrastructure is needed. Traditional databases are designed for tables and documents rather than vectors, which makes them inefficient for machine learning models. And most companies lack the expertise to build a dedicated data infrastructure themselves. As a result, Pinecone says, developers usually have to compromise between speed, accuracy, stability and scale when running machine learning models.

The answer, according to Pinecone, is its vector database. Pinecone says it can dynamically transform and index billions of high-dimensional vectors to answer queries such as the nearest neighbor and max-dot-product search extremely accurately in just milliseconds.

Liberty told SiliconANGLE that Pinecone’s database was inspired by the in-house systems that big tech firms such as Amazon and Facebook Inc. use for vector-based search. But he noted that those systems are highly specialized for the specific products they support.

“Pinecone, on the other hand, has developed a proprietary vector index that is highly accurate and efficient across all applications,” Liberty said. “We also provide a container distribution layer that scales to accommodate any workload, and a cloud management system that lets anyone onboard in minutes and start deploying ML applications to production.”

Liberty added that Pinecone’s vector database is quite different from similarly-named databases that have appeared before.

“For example, Actian sold a ‘vector database,’ but only in the sense that it used vectorized instruction sets,” Liberty said. “Other than that, it was a standard SQL database. Pinecone, in contrast, is a database for numerical vectors.”

The Pinecone database can support dozens of popular real-time AI use cases, including personalization, semantic text search, image retrieval, data fusion, deduplication, recommendation and anomaly detection. And its serverless nature means that customers don’t need to worry about managing computing resources or maintenance as the platform can scale out as it’s required.

Constellation Research Inc. analyst Holger Mueller told SiliconANGLE that Pinecone could be onto something big with its vector database, since it’s well-known that traditional databases are inefficient when it comes to AI.

“This could lead to the creation of a new database category,” Mueller said. “But before that happens, we will need to see more adoption, and probably more vendors releasing and supporting vector databases designed with machine learning as their core competence.”

Image: Pinecone