Artificial intelligence-driven search and discovery startup Marqo.ai said today it has closed on an A$8 million (about $5.2 million USD) seed funding round.

The lead investor in the company, officially known as S2Search Australia Pty Ltd., was Blackbird Ventures, with Creator Fund, January Capital and the co-founders of Cohere Inc., Ivan Zhang and Aidan Gomez, also participating.

Marqo is trying to upend the world of search, which has traditionally always been keyword-based. Instead of using keywords to surface its results, the company says, it relies on cutting-edge machine learning models that use vector search techniques to return more accurate findings by understanding both the searchers’ content and meaning.

It enables users to search using images, text or a combination of both. The technology is intended to power search features for e-commerce sites and other marketplaces, but can also be applied to areas such as generative AI, data analytics and security, the company said.

Marqo explained that its technology is a new form of vector search that continuously improves by learning from user engagement, differentiating it from existing vector databases. Vector search is often said to be better suited to finding unstructured data such as documents, videos and images, because they can be converted into vectors, or complex numbers that AI algorithms can index more easily.

Unlike traditional vector databases, Marqo includes both storage and inference out of the box. Users can store and query unstructured data such as text and images through a single application programming interface. It integrates easily with generative AI technologies and acts as a knowledge store for large language models such as OpenAI LP’s GPT models and open-source commercial use models such as DALL-E, OpenLLaMA and others.

Marqo co-founder Tom Hamer (pictured, right) said the search industry is desperately in need of modernization, as the vast majority of search experiences today are still powered by keyword-based systems that provide substandard results. “Customers want search experiences that anticipate their needs, not just match keywords,” he stressed. “This is especially important for businesses where the search bar is the core product, such as ecommerce. By automatically improving based on user interaction, Marqo provides highly relevant results, and increases customer conversion rate, order value and revenue.”

Vector search is known to be superior for many search use cases, but the problem for most companies is finding a way to implement the technology, and that’s where Marqo comes in. “Vector search remains challenging to implement especially for applications requiring real-time search,” said Marqo’s other co-founder, Jesse Clark (left). “[But] it is rapidly becoming a must-have for generative AI applications. We want anyone to be able to leverage the latest machine learning models, even if they’re not an expert in this field.”

Marqo has made the code behind its search technology open source, so anyone who wants to can use it for free. In addition, it’s offering a fully managed and serverless cloud service for those that require a plug-and-play version without any operational overheads. The startup said this service is now generally available, following a successful early access program. It added that its continuous-learning functionality will be launched later this year as part of the Marqo Cloud service.

Cohere’s Gomez said he believes Marqo can be instrumental in making AI more useful to businesses, especially generative AI technologies. “Vector search is a huge growth area,” he said. “[It] has become a must-have component of generative AI.”

Blackbird Partner Nick Crocker pointed out that the vast majority of the 100 trillion gigabytes of data in the world is unstructured, meaning it cannot easily be accessed by traditional keyword search engines. “Computers and companies need new ways to search,” he added. “Marqo represents the next generation of search – hyper-relevant, AI-powered and based on human understanding.”

Photos: Marqo.ai