By clicking “Accept All Cookies”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.

Brightwave: An AI Investment Research Assistant For Everyone

We are excited to announce our investment in Brightwave, the AI research assistant purpose-built for investment professionals. Generative AI has been one of the fastest growing technologies in recent years, permeating every field from developer tools to legal. The investment world has been one of the pioneers in applied machine learning, starting with groups like James Simons’ RenTech in the 80s to today’s quant funds. And yet, the applications of generative AI for investment professionals so far have been limited to increased productivity: summarizing documents, asking questions over SEC filings, etc. But the real value investors get is not from facts, but insights - how does this new information fit into my worldview, and how should I change my strategy accordingly?

Brightwave is an AI research assistant that generates trustworthy, insightful financial analysis on any subject. It helps users reason about risks and opportunities within specific market and sectors, understand how different industries influence each other, drawing links between disparate topics such as rare earth metal processing and and GPU production, and can provide sentence-level attribution for every research claim so that the user can investigate further either manually or through the Brightwave chat. 

Mike Conover and Brandon Kotara launched Brightwave 4 months ago, and their customers represent over $120B of assets under management, ranging from owner-operated RIAs to $20B+ crossover hedge funds. We talked with Mike in our founder Q&A to explain more of the story behind Brightwave, and if you’re interested in the more technical details, Mike was also a guest on today’s Latent Space Podcast by our partner Alessio. 

Where did you grow up and how did you find out about computers?

I grew up in a small mountain town and discovered the power of computers programming simple video games at an early age.

How did you and Brandon meet and what made you both decide to start Brightwave?

Brandon and I met working together on sister machine learning teams at Workday, where I was leading teams of engineers focused on automation in the Office of the CFO while he architected 100M-scale semantic search systems for the human capital management product line.  We’ve both been long-fascinated by markets; Brandon is the former CTO of a federally-regulated derivatives exchange and clearinghouse, while I’ve authored papers (Nature, 2018) showing how 500M LinkedIn job transitions are predictive of next-quarter S&P500 market cap changes.

For myself, I believe that when humans interact with technology they create digital trace data in a process similar to the way social insects like ants and bees imbue their environment with chemical signals to coordinate collective action. Language models provide a technological foundation for building compound systems that can reason over these signals at planetary-scale – the closer we looked the more it became clear that this kind of autonomous investigative analysis would be a major unlock for the work of analysts trying to understand the global financial system.

How is Brightwave going to change the lives of the professionals you serve?

At Brightwave, we're deploying AI that expands people's ability to reason about markets, sectors and the economy in ways that exceed the limits of human cognition. Whether it’s an analyst who needs to understand something about the market nobody else has seen, or a professional who needs to lead with a sophisticated view on a broad range of topics, connecting the dots across huge volumes of content is an extremely difficult task for humans, but is an area where compound systems built around language models really shine.

Our view is that these technologies should aim to be a "partner in thought", augmenting humans rather than automating away the most interesting and meaningful parts of the job. The comparison I draw is to finance professionals in the 1970’s prior to the advent of computational spreadsheets. Step back in time and you see a lot of rote, manual calculation – in 2024 accounts are still an extremely important part of the economy, but the sophistication of the analysis they can bring to bear on any problem has scaled far beyond what’s possible doing the work by hand.

How are customers using the product today, and what has the initial response looked like?

The response from the market has exceeded our expectations. From half-billion dollar RIAs to crossover hedge funds managing tens of billions of dollars, the product is resonating. Our customers represent over $120B of assets under management and that number is growing all the time.

For larger funds, the research process and developing conviction in a thesis is key. Brightwave enables users to go deep and see around corners, not just summarizing facts but synthesizing insights across a universe of content, contextualizing fact patterns and teasing out second and third-order implications that humans are likely to miss.

We’re also seeing broad-based demand from professionals who need to expand their coverage or get up to speed quickly on new topics, equities or sectors. Whether it’s wealth managers who need to provide bespoke market insight to large client bases, corporate strategy teams looking to understand and counterposition during earnings, or analysts looking to expand their universe of coverage, Brightwave allows these professionals to rip through filings, transcripts, news, sell-side analysis, and more to develop comprehensive market views extremely quickly.

The real value is not just facts, but insights - how does this new information fit into my worldview and strategy? What does it entail? That reasoning and second-order thinking is where Brightwave shines.

What are some of the key challenges in building an AI system for financial analysis? How do you approach them?

Rather than fully open-ended retrieval and reasoning, we decompose the problem into subsystems with specific goals - it’s a matter of orchestration, and this is a place where we’re uniquely positioned to build something very powerful owing to the team’s background in machine learning and artificial intelligence. It reflects a distributed systems mindset; we train and build model subsystems that perform very specific behaviors, and harness them together in a way that allows us to experiment and evaluate incremental improvements to the system’s planning and reasoning capability in a very controlled way. We’re realizing a fairly high degree of engineering and methodological leverage, and this allows us to run experiments with very high velocity and in a robust, end-to-end way.  

You see a concrete example of this in our information retrieval systems. You have to deeply understand a user’s intent and identify the most relevant set of passages and documents that are uniquely appropriate to a specific instruction. Even with the advent of large context windows it’s very much like a bin packing problem, and the naive approach of simply embedding or using classical retrieval methods just isn’t going to make the cut when you’re dealing with sophisticated end-users.

What is your long term vision for Brightwave?

We’re on a mission to transform the way humans understand the world, starting with the global financial system. I believe we are creating a system that is capable of extending the limits of human cognition, empowering us to reason about and act upon the planet-scale structures that govern the course of human society. It’s an extremely ambitious project, but we have an exceptional team – you’ve got NeurIPS-published research engineers working alongside talent from Meta and Goldman Sachs – we’re positioned to change the world and intend to do exactly that.