Investment Notes: Marqo

We are pleased to announce that Blackbird has led the Seed round for Marqo. 

We first invested in Marqo’s pre-seed round in May 2022 and are excited to announce them to the world!

We are joined in the round by January Capital, The Creator Fund, and several angel investors including the VP of AI at Replit, Michele Catasta, and Cohere Co-Founders, Aidan Gomez and Ivan Zhang.

‍

Search the way you think

Marqo’s vision is to elevate the performance of search to the level of human thought.

Search is everywhere. Just think of the number of times today that you have searched on Google, an ecommerce website, a news site, or your work drive.

Search is also a critical piece of infrastructure. Let’s take Amazon and say you were able to drive a 1% improvement in the search results on its online marketplace, measured by click-through conversion. That would add $4B in revenue to their annual statements.

And Marqo is already driving higher rates of improvement in search than that for customers today.

How? Marqo gives traditional keyword search and other search models an AI-facelift. 

Marqo’s technology is underpinned by vector search, a computational technique that facilitates effective information retrieval from large and multimodal data sets. 

Marqo uses machine learning to first transform text, images, video, and audio into numbers encoded with different characteristics and then bundles them into multi-dimensional objects called “tensors”. By doing this, the model can directly compare, say, a picture of a dog with an audio clip of a dog barking.

Vector databases have become popular of late with the rise of generative AI and Large Language Models (LLMs), which open up numerous additional applications of the technology. The performance of LLMs directly correlates to the quality and structuring of the underlying data it queries. This is where vector databases come in.

But Marqo is so much more than just a “vector database”. 

First, it is built for scale and real-time search. For most search use cases, speed is important and this is where Marqo excels. 

Second, it is built with user centricity in mind. Marqo bundles so much technology and control into one simple api, abstracting away the complexity. In three lines of code, developers are incorporating human-like understanding into their applications.

There is over 100 trillion gigabytes of data in the world, the vast majority of which is unstructured. 

We believe that every search bar will use vector search and Marqo is on a mission to capture as many of these as possible.

‍

Returning home

Marqo is founded by Tom Hamer and Jesse Clark.

At age 14, Tom taught himself to code and built his first iPhone app - an app that, ironically, taught other developers how to create their own iPhone apps! It quickly attracted over 40K users. Tom went on to study computer science at ANU before undertaking a Masters at Cambridge University specialising in machine learning. Most recently, Tom was an engineer at AWS working on their Relational Database Service. 

Jesse is a physicist turned machine learning expert. He spent a number of years as a postdoctoral researcher at the University College London then as a research fellow at Stanford University. He was the principal ML researcher at Stitch Fix before leading ML in Amazon’s Robotics division. 

Both Tom and Jesse returned to Australia, from the UK and USA respectively, to build Marqo from its present HQ in Melbourne.

This is one of the more impressive technical founding teams we’ve met. With Marqo, they are set on doing their life’s work. They are magnets for talent and we are impressed with the team they are building around them.

‍

Democratisation of technology

Marqo believes in the democratisation of technology.

It has taken an open source approach and is rapidly building a devoted developer community across its github, slack, and community forum. 

In venture, we talk about the “signals of customer love”. 

One Marqo customer runs a major ecommerce website with millions of item listings. Their challenge was in orchestrating a more effective search experience that navigated errors in metadata (eg. a seller lists an item as a ‘green shirt’ when it is actually black) and complex unstructured data in multiple formats. 

The customer ran a number of sitewide experiments where Marqo was handling millions of queries per day, and the results were so compelling that they are now rolling the upgrades into full production.

That is strong signal.

We love how deep this ecommerce vertical is. Ultimately, Marqo’s mission is to be the leading end-to-end neural search solution for all applications.

We are excited to be investing in them right at the beginning.

Ps. if you’d like to read more on Marqo, you can read profiles of their Seed round in Techcrunch and the AFR