Inside 8 innovative technology projects on Wall Street, from Blackstone to Goldman Sachs

Project name: Deep Research

What it is: A generative-AI tool to answer complex research questions and automate the bulk of junior analysts' work

Lead executive: Chen Fang, data and analytics lead on Balyasny's Applied AI team

Balyasny Asset Management is a step closer to its goal of building an AI equivalent of an analyst thanks to a new tool called Deep Research.

Built by the hedge fund's Applied AI team, Deep Research helps analysts and portfolio managers answer complex research questions. The tool pulls in info from about 5 million documents, like regulatory filings, earnings transcripts, third-party research and market data, and Balyasny's internal analyses and memos. Mostly used by investment teams, Deep Research helps analysts and PMs research stocks before making a trade and gauge the impact of global market events on a portfolio or set of stocks.

It's in beta and being tested across roughly 50 teams. Those teams send their questions to the Applied AI team, which passes them on to the bot. The goal is to release the tool firmwide by the fourth quarter, with every team accessing the tool directly.

Fang said the firm wanted to automate tasks for analysts to "reduce their research process from days and weeks to minutes and hours."

In one recent example, a PM asked Deep Research to find companies whose supply chains were affected by tariffs. The tool scanned more than 20,000 documents to identify 120 companies and provide a report with explanations for each company — all in about an hour.

Going forward, Fang said the team aims to offer PMs more actionable insights and trade ideas that go beyond summarizing and linking back to documents.

Project name: DocAI

What it is: Building a generative AI tool that crowdsources important documents for search and summarization in a curated way.

Lead executive: John Stecher, chief technology officer

The investing giant Blackstone has spent the past 10 months building DocAI, what Stecher called "the next evolution" in the firm's generative-AI journey.

Workers across Blackstone will be able to upload documents such as confidential information about specific deals and macroeconomic research from investment banks and consulting firms. The idea is for employees to ask DocAI for information within those documents and for the tool to then find and summarize it.

"While it's great that ChatGPT and whatnot allows you to search the broader internet, a lot of times what really matters is asking questions and getting summarizations of very specific topical documentation," Stecher told BI.

Some groups at the firm are using DocAI, and Blackstone hopes to this fall start making it available to its 5,000-strong workforce.

Stecher said analysts and vice presidents in real estate and private equity were using DocAI to research potential deals. Legal and compliance teams have used it to keep track of and query different credit covenants and mandates. And thanks to architecture diagrams they've uploaded, software engineers can better understand the connections between different systems.

"We don't just treat this as 'dump a bunch of information in,''' Stecher said. "Instead it's 'let's make our platform a highly curated set of information that individuals select to upload and view as valuable to Blackstone's operations.'"

Project name: Citi Integrated Digital Assets Platform

What it is: A blockchain platform for the bank to offer institutional clients digital-asset products

Lead executive: Nimrod Barak, head of Citi Innovation Labs

Despite crypto's upturns and downturns on Main Street among retail investors, a top tech exec at Citi says there's still value and potential for the underlying technology on Wall Street.

The bank launched CIDAP this summer for its blockchain offerings for institutional clients. The aim is to consolidate Citi's digital-asset products and services, which span liquidity management, trade finance, bonds, and custody. Doing so not only makes it easier for the bank to manage and update its offerings but could make it faster to onboard any new use cases, Barak told BI.

Citi offers several blockchain-based services. One is tokenizing cash to quickly move deposits between Citi branches globally at any hour. Another is digital custody and settlement services to speed up processing times.

Barak said that blockchain was maturing, now able to be used for enterprise infrastructure and institutions. He added that corporate clients were increasingly interested in the technology.

"We see that blockchain has emerged as a materially positive impact on financial services across the value chain," Barak said.

Project names: Assistants, LLM Gateway, DocLab

What it is: Building blocks for programmers to develop customized generative-AI tools

Lead executive: Neil Katz, managing director

While most Wall Street firms have opted for a generative-AI strategy wherein a core tech team builds ready-to-use tools for the whole firm, the quant giant D.E. Shaw has spent the past year developing ways for developers and investment teams to build their own tools.

"I often describe our company as a set of business units that innovate independently but seek opportunities to collaborate," Katz, who oversees much of the firm's quant-technology development, told BI. "Individual business units often have their own technologists so teams can pursue experiments and make local decisions that are optimal for their group."

D.E. Shaw's generative-AI suite has three main components. Assistants, an AI chatbot, was rolled out earlier this year. LLM Gateway lets employees access about two dozen external large language models, the tech behind generative-AI applications like ChatGPT. Finally, there's DocLab, a database library of millions of external documents, like regulatory filings and news articles, that can be queried for summaries or specific info.

D.E. Shaw wants to empower its developers to create generative-AI tools specific to trading desks and investment groups, integrating their own data and software systems. In some cases, with as little as 10 lines of code, quant researchers can connect to an AI model through LLM Gateway and have it consult on their research.

Project name: Legend

What it is: Creating one platform for all the bank's data to automate more tasks

Lead executive: Neema Raphael, chief data officer

At Goldman Sachs, every dollar invested, trade executed, company met, and loan financed is another data point that could fuel the bank's analytics engines.

But as it amassed large troves of data, less time was spent organizing that data and figuring out how different teams would access it. That's why Goldman Sachs developed Legend about a decade ago to be its one place for accessing all its important data.

Doing so could allow employees to quickly unearth connections that could lead to multimillion-dollar deals, automate some operations-heavy work, and let quants and data scientists build AI models that can find new patterns.

Now Legend has become a priority, thanks to the explosion of data — and its potential to train generative-AI models and automate back-end processes.

"The most important things that we do, every business function and every business process, has some touchpoint to Legend and Legend data," Raphael told BI.

Legend allows everyone at Goldman — software developers, risk-management experts, and bankers — to see and use the same set of data, giving the bank "one version of the truth for all use cases," Raphael said. Through that consolidation, Goldman can also save on infrastructure and operational costs, since data doesn't need to constantly be reconciled, processed, and copied from system to system.

Raphael said the bank wants to use Legend to help grow its asset- and wealth-management business and its sales and trading business, where data issues could hinder automation for certain processes.

"What we've noticed is that at some tipping point, the automation and scale falls down a bit," he said, referring to back-end processes, "and it's usually because of data issues that you couldn't automate."

Project name: RealHouse

What it is: A way to analyze and conceptualize KKR's real-estate credit and equity businesses

Lead executives: Jessica Ciaccia, a director on KKR's real-estate team, and Jon Knehr and Steve Lo, directors on KKR's technology team

In less than 10 years, KKR's real-estate credit business has grown from $1 billion to over $37 billion in assets under management. Through this unit, KKR invests in debt opportunities by extending loans and other credit solutions to real-estate operators, developers, and owners and by purchasing commercial mortgage-backed securities.

To keep up with the growth, KKR's technology and real-estate teams came together to build RealHouse. The platform is helping the Wall Street giant analyze and conceptualize its portfolio of more than 1,000 investments.

More than 200 employees, from business leaders and originators to analysts and operations teams, use RealHouse to see deal overviews, performance metrics, and related documentation, as well as fund composition and risk exposures, all in one place.

Before RealHouse, a portfolio manager wanting to assess the risk of financing a multifamily property in Miami would have spent hours assembling data from several sources. Now this information can be pulled almost instantly. RealHouse also automates some routine tasks, like monitoring interest-rate caps and generating funding projections.

Ciaccia worked closely with Lo, a director of software engineering, and Knehr, another technology director, to build the platform. The team plans to expand the tool to the firm's real-estate private-equity business.

KKR has a stake in Business Insider's parent company, Axel Springer.

Project name: Condor

What it is: Rebuilding a key platform for Man Group's systematic-trading business to take on more data and asset classes

Lead executive: Barry Fitzgerald, cohead of front-office engineering

The world's largest publicly listed hedge fund is in a multiyear rebuild of its systematic-trading and quantitative-research platform.

The new platform, called Condor, was initially designed to serve Man AHL, the firm's systematic-investing arm. But Fitzgerald has bigger plans: more asset classes, more investing styles, and more data.

"The worst thing we could do with this is build something that fulfills our need for exactly today," Fitzgerald told BI, adding: "We don't know how we will trade in two years' time, but I would hope this platform runs for 10 years or longer. It is a big multiyear project, so it should get the payback."

Fifteen years ago, Man AHL was known mostly for its systematic futures trading, but the firm has expanded to other asset classes, like equities, corporate bonds, and options, with varying holding periods. Over time, Man built different systems to trade them all.

With Condor, Fitzgerald aims to bring those together — the research, the trading, and the logging — into a single platform. Because Condor will replace what is now a bunch of different systems, a variety of workers will use it: Quant researchers could develop investing strategies, tech teams could add features, and risk and operations teams could support trading systems. He said that having one platform for all these uses would help provide a cross-asset view of its risk exposure, conceptualize allocations across all assets, and do more analytics.

Work on Condor began about 18 months ago, and Fitzgerald expects it'll be another two or three years until it's fully integrated. Quantitative researchers are already using it to experiment with statistical models. He added that some calculations for big multiasset research graphs now take 30 minutes instead of 12 hours. Over time, he said, it'll hopefully expand to encompass all of Man's asset classes and trading styles.

Project name: Lightning

What it is: A firmwide data platform designed to quickly move and process data, saving engineers time and speeding up analytics

Lead executive: Mona Eldam, head of technology in Singapore and distinguished engineer

When Morgan Stanley began migrating to the public cloud in 2021 from its on-premise data centers, it needed something that could help the bank move its data quickly, handle loads of different types of data, and ensure that the data quality didn't falter during the transfer.

Eldam rounded up her troops of data engineers spread out across 10 locations globally. Her team is responsible for managing data on behalf of the bank's investment and wealth-management businesses as well as its institutional securities unit.

In those early days, Eldam had to solve for moving about 20 sources of data, numbering hundreds of entries, every 10 minutes.

Lightning, which was initially built to move and bring together these disparate types of data, has become the standard for how Morgan Stanley moves data, regardless of whether it's on-premise or the cloud. Lightning underpins more than 80 client applications. It moves several terabytes of data for Morgan Stanley every day, and the firm says overall data-migration turnaround time has been reduced by 75% on average. Thanks to Lightning, engineers don't have to spend time developing their own frameworks for moving data.

Eldam said her team was adding functions to Lightning, including the ability to handle data in charts and graphs and ways to flag when a job needs more cloud capacity.