Sam Green [0:12]:
Good morning, everyone. I'm Sam Green, team lead for Applied AI at Bridgewater
Associates, and this is my colleague Aaron Linsky. Today, we're here to talk to
you about the Vault, the data platform we've built at Bridgewater to turn
unstructured data into a first-class asset in our investment process. As part of
the talk this morning, we'd like to share three lessons we've learned building
the Vault. The first is that humans and agents have very different needs from a
data platform, so you have to build for those different needs separately. The
second is that the greatest impact occurs when a single tool can meet both of
these sets of needs at the same time. The third is that the real prize is when
you can build the foundations needed to make agents a proactive participant in
your investment process or whatever you're doing with your tools, instead of a
reactive tool for your humans to use.
Before I dive deeper into talking about the Vault, I wanted to start with some background about Bridgewater. Over the last 50 years, Bridgewater has been in the constant pursuit of the world's best understanding of how markets and economies work and turning that into a systematic encoding that we can use to stress test our logic over time. What originally began as a set of if-then rules on a yellow legal pad evolved into a continuous compounding process with humans at the center, constantly looking for the next thing that's changing in markets. More recently, we've been working on building AIA, the artificial investment associate, which puts a machine at the center of this continuous process.
Across all these phases of evolution, we've had a constant philosophy in mind: we want to be constantly looking at how the world is changing and what's going on in the world, noticing what dynamics are not yet encoded in our understanding, and using that learning to improve and compound for the future. For most of Bridgewater's history, that meant building and overseeing time series models on structured data. That structured data needed to be carefully built so that it could be safe for use in our research process and aligned to our internal ontology and world model, so we could get great leverage out of using the data we had built.
With the emergence of language models, we had a huge new opportunity to engage with a much wider world of data—news and research, transcripts, policy documents, and all of the internal conversation that was previously just in the ether, such as internal memos and meeting transcripts and debates happening inside the firm. We quickly realized that all of that data needed the exact same treatment to be fit for purpose. It needed to be carefully cleaned, governed, and made ready for use in both point-in-time research questions today and also our systematic study over time.
When we looked out at the landscape of data vendors and data platform providers, there was nothing that provided us with all of the control we needed to build and maintain our data advantage. So we built the Vault, which is our repository of external and internal unstructured data, and the platform that indexes, organizes, and makes all of it ready for use by our human investors and agents. Today, the Vault has more than 100 million documents in it, with thousands more added every day, minute by minute, from 135 external sources, and again, growing all the time, as well as all of the data from inside of Bridgewater. Crucially, all that data is not just dumped on our investors' desks; it is carefully aligned to our internal ontology with enrichments like the country's concepts, market, and most importantly, the time period discussed in the document.
To bring this to life, I want to start with a quick motivating example. I asked a fairly basic representative investment question to an enterprise chat client: how is AI infrastructure investment impacting the Korean won, and do market participants expect this to continue? As you're all, I'm sure, familiar with, the agent goes to the web and looks for resources it can draw on because I've explicitly asked it to find other people's perspectives on the question. It zooms in and finds 15 relevant-looking resources to draw on before it gives me back its answer. When I zoomed in on the answer it provided, a couple of things stood out to me. The first is that even though it found 15 relevant-looking sources, it only picked one or two to answer my question. The second is down at the end of this response, where the agent itself acknowledges that it hasn't answered my question. It didn't find enough documents on the open web to establish a market consensus and meet the parameters I laid out.
When I double-clicked even further, the problem got even worse. One of the notes that it primarily drew on to answer my question was published way back in July of 2026, even though I asked this question and captured this answer just at the end of last week in early October. In the age of AI and in investing, three months is a massive amount of time. It's not even close to good enough to draw on this resource to answer my question. When I ask this exact same query attached to the Vault, something totally different happens. The first is that the connection to the Vault helps to prompt the agent that it should look for a specific period of time when answering my question. So it zooms in on October 1st to October 8th. It finds a much wider array of resources—52 candidate documents that are relevant to the query over this specific period of time—because of the Vault's much more significant corpus of content that's relevant to investing questions. When it ultimately answers my question, it draws on seven instead of only one or two and gives me back a rich answer that the agent is more confident will address my question.
For us, this is the big lesson from building the Vault to motivate the talk today: it takes great data and the right connectors to make agents impactful. I'm going to hand over to Aaron now to talk about some of the product engineering takeaways from building the Vault.
Aaron Linsky [6:53]:
Thanks, Sam.
Aaron Linsky [6:55]:
Hi, I'm Aaron. I run product engineering at Bridgewater. I'm going to take us
through a few of our learnings from building the Vault over the past few years.
I'm going to start with one that might seem obvious, but I still think it's
worth saying: search is hard. But it gets even harder if we're designing for two
different consumers—in this case, humans and agents. I'm going to go over the
difference here, starting with us. Humans need precision and a high
signal-to-noise ratio because we can only read so much. Even a fast reader
manages about 300 words a minute, and personally, I'm much slower. Agents are a
different kind of reader. They're far less sensitive to precision. If the data
is there, they'll find it, and basically anything you hand to them will get
attended to. They scale super quickly and consume about 100 times faster—30,000
words per minute. So we need to build the Vault for both of these.
When we're thinking about the dimensions that matter, there are five on the board: recall, precision, filtering, entitlements, and scale. Some of them matter to people, some matter more to agents, and many to both. I'm going to go deep on three of these: precision for people, scale for agents, and then filtering for both.
So, precision. Once we filled the Vault—which was a heroic effort by our data and legal teams—recall was no longer a pressing issue. However, the number one complaint was that Vault search is bad. As the product owner, it's painful to hear that, but my response was, let's benchmark it. That's what we did. We collected questions and found the best documents by hand, including ones that looked helpful but weren't. In total, it was about 200 examples, and it took us less than a week. It was not fun or very glamorous, but it was very helpful. After we ran it, it was very clear: search was bad. The key metric we were looking at is precision at 10, which is of the top 10 results, how many are the right documents, and we were hitting about 60%. So the people were right; it was bad. But now we had a way to measure all of the improvements, and here on the board are five of them. By the time we got done with Agentic Search, we were up in the 90-95% recall at 10 or precision at 10.
Now on to filtering. Like anyone, we want the most relevant documents. But when we looked at the production traces, we saw that almost always there was some condition—either as of a certain date, from a certain source, or about a specific asset. All of that requires advanced filtering, and that's where we ran into our first major performance issue. There are classically two ways to filter a vector database. Post-filtering will use that vector index first and then filter the top results by metadata. It's fast, but recall could be pretty low because the globally relevant results might not match your filters. On the other side, we have pre-filtering, where we'll filter first and then calculate distances for every one of those matches. The advantage there is that the recall is very good, but it can be very slow, and we needed something in between. That led us to try and ultimately switch to turbopuffer. Their approach, which they call native filtering, looks only in the nearest clusters that contain matches and then scores just those. What you end up getting is something that's fast like post-filtering, but with close to the same recall as pre-filtering.
Lastly, I'm going to talk about scale. Human scale for us is pretty pedestrian—100 queries an hour. We have 300 users who are going to search every so often. It takes our users minutes to read the documents they're receiving. Agents, on the other hand, are a few orders of magnitude beyond that—easily 100-plus queries a second, and agent forms can hit thousands easily. They obviously will read, as I said earlier, documents in seconds. So we needed something that could scale as well. Again, turbopuffer handled that easily too. We were covering all five dimensions. I didn't talk about entitlements, but I'll be discussing that later today. We covered it all, and it unlocked new workflows.
However, it didn't change how our researchers worked. They still weren't coming to the Vault every morning. My second lesson is that no matter how amazing the tools are, we can't just drop them on our investors' desks and expect them to be used. We need to migrate their old habits into a new form. We looked at what they were already doing. Many of them were living in email. The research came straight from the brokers; they had subscribed lists and rules that filtered everything into folders. It worked, it was familiar, and they knew they weren't missing anything. So we had to build all of these features into the Vault. That's exactly what we did. Everything's searchable; the subscribed lists became saved searches, and folders became saved collections. It's the same habits, just a new home. Once we had them there, we could open up new habits. They can ask questions across all the data, as Sam showed. They can follow dynamic feeds by topic, and they can hand those collections over to their agents to take further action. That's when we clearly had product-market fit. Once we had that fit, the users wanted it everywhere, so we put it in chat, made a phone app, integrated it into Slack and their IDE, and of course, we availed all of that to the agents as well, including a CLI because agents love CLIs.
The main question here is, okay, so what does this enable? Sam's going to talk about what we're building right now, but first and foremost, it's helping us to move around that research circle faster. With our pocket analyst, which is our first AI analyst, it can do hours of research in just a few minutes for a few dollars, and that's relying on the Vault. We're expanding that now with the researcher, which is the next generation. It's the same idea stretched across days or weeks. We're also able to train custom models to more intelligently reason over the Vault's data and use those together with other foundation models to feed systematic decision-making processes. The forecaster is our first one, which produces AI-generated macro-relevant predictions. We've also built investment signals from the Vault, turning unstructured data into structured, tradable strategies.
Now I'll hand it back to Sam to take us through the Vault's latest capability.
Sam Green [14:07]:
Like I alluded to at the beginning of the talk, the real prize is when you can
take agents from a reactive resource that can be ready to answer a human's
question and instead make them proactive participants, helping to steer your
high-value professionals toward the most important areas for their attention.
We're solving that problem in an experience we're calling the Investor
Workbench, where the agent is always on, tracking what's important to that
investor, and presenting them with the areas that are most in need of their
attention. It looks like an experience that relies on the bot, where a topic is
proposed and several pieces of information are presented to the investor to
decide whether they should dig deeper or ignore that topic entirely. You can see
here the card that an investor sees to decide whether to dig into the same AI
impact on currencies question we looked at earlier in the talk. The investor
receives everything they need: an agentic summary if they want a quick scan
across the agent's findings, the exact evidence that the agent was drawing on so
they can assess for themselves what was leading to this recommendation, and if
they choose to dig deeper, they have a direct connection to the research output
from the analytical tools that Aaron just described.
When they dig deeper alongside the agent to look at the dynamics driving this question, the learnings are automatically captured. The results of their collaboration are automatically captured back in the Vault. For us, this is the key lesson: the Vault enables compounding because it provides the right evidence, the right internal and external information at the fingertips of our investors and agents, and it automatically captures the synthesis they generate and puts it back in the corpus for the next turn. This makes us go around the circle faster and better. When we think back to the question we began with—what it takes to make an agent work—it’s the right context and the right understanding of what's going on in your world. Our question for you is: does your AI know what's going on in the world, and does it remember what your firm has learned about it? Because we are teaching ours too. If these sound like an exciting challenge, we'd love to talk. Thank you.
