DATA

How AI Maps Company Connections

Researchers from Domyn used AI agents to build a knowledge graph from SEC filings.

If I say ‘Microsoft Excel,’ you likely think of a spreadsheet of rows and columns.

But if I say ‘knowledge graph,’ you probably can’t picture it. That’s because until recently, knowledge graphs and other graph databases were hard to build at scale.

Knowledge graphs have been around for decades. They’re typically made up of “triples,” such as Matt (1) writes (2) AI Street (3). They’re pretty straightforward:

• Matt writes AI Street

• AI Street is a newsletter publication

• AI Street covers AI adoption in financial services

• Matt lives in Milan

• Matt worked at Bloomberg News

• Bloomberg News is part of Bloomberg LP

• Matt is from Rhode Island

You can see how the idea is simple, but even small graphs become unwieldy as you add more connections.

Manually building them is tedious and costly, but AI is making the process far more efficient.

LLMs can extract entities and relationships directly from unstructured text without needing massive labeled datasets or hand-built rules. Instead of a team of analysts tagging ‘Walmart’ as a company or connecting it to ‘supply chain disruptions in Vietnam,’ a model can infer those links. They can also convert earnings calls, SEC filings, or news articles into graph updates with much less friction. (They don’t do this perfectly, but you get the idea.)

This framework makes answering previously tough questions easier, such as:

• What exposure does this foreign subsidiary have to tariffs in Thailand?

• How many other joint ventures has ABC entity announced?

• Which subsidiaries or SPVs are ultimately owned by XYZ?

Researchers at Domyn built FinReflectKG, an open-source financial knowledge graph from SEC 10-K filings of S&P 100 companies. Since the data is not labeled, their system uses AI agents that iteratively refine extracted information through feedback loops, improving the quality of relationships and entities captured. In testing, the approach achieved about 65% accuracy on rule-based evaluations, so not yet ready for primetime. To encourage further research, Domyn open-sourced their dataset — all 17.5 million “triples” — on HuggingFace.

What I like about knowledge graphs is that they help answer broad-based questions. I'm biased as a journalist, but the right unanswered question can unravel the whole story.

Takeaway

AI makes it possible to build knowledge graphs at scale. There are still some kinks to work out, but I wouldn’t be surprised if, in a few years, knowledge graphs approach Excel’s ubiquity.

KGs + Wall Street

If you’re building in the knowledge graph and investing space, please reach out: [email protected]. I’d like to learn more about how folks are using them.

NEWS

BlackRock Brings AI to Advisors’ Desks

BlackRock rolled out an AI tool in its Aladdin Wealth Platform that helps financial advisors understand what’s driving changes to their clients’ portfolios. Morgan Stanley, which has thousands of advisors, is the first to adopt.

The tool draws on Aladdin’s analytics, each firm’s CIO outlook, and client portfolio data to generate short, plain-language summaries of what’s driving performance and risk.

I asked BlackRock how it handles hallucinations since that’s likely the biggest concern from a money manager. Here’s what Matt Sahn, Head of Product & Engineering for Aladdin Wealth Tech, said through a spokesperson:

• “The system has passed rigorous validation by BlackRock and major clients,” Sahn said. “We use a combination of ground-truth testing, human review, and sampling to measure accuracy. And since commentary is advisor-facing only, there’s always a human reviewing the output before sharing it with clients.”

What makes this different from other advisor tools, he added, is its integration into Aladdin Wealth.

• “It can securely pull in client portfolio data, firm market views, and Aladdin analytics — and it’s customizable for each firm,” he said.

Takeaway

We’re still very early in AI and investing adoption, but this is one of the first GenAI tools from a major firm designed for use outside its own walls. Tying it to advisors is smart—there’s always a human in the loop.

NEWS

AI Boosts AlphaSense’s ARR to $500M

AlphaSense, an AI-powered market intelligence platform, has lifted its annual recurring revenue to $500 million, up from $400 million in March this year and $200 million in April 2024—driven in part by rapid AI adoption among financial institutions and corporations, the company announced this week.

The company’s AI agents now automate research workflows. One of the company’s agents can autonomously conduct expert interviews. Users can select a topic, review AI-generated questions, and receive a transcribed expert interview faster than scheduling and conducting traditional human-led interviews.

The technology helps shrink the time it takes to conduct "channel checks," where analysts interview multiple industry sources to understand real-time market conditions like pricing trends, supply and demand shifts, and competitive dynamics.

The company says it’s not replacing human-led expert calls, which remain central to investment research. Instead, the AI handles overflow work, explores emerging topics where research is thin, and can operate around the clock, according to a company spokesperson.

AlphaSense has more than 6,500 clients, including 88% of the S&P 100.

Takeaway

The more AI agents get embedded into business, the more they shift from tools that summarize information to systems that generate it.

SPONSORSHIPS

Reach Wall Street’s AI Decision-Makers

Advertise on AI Street to reach a highly engaged audience of decision-makers at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email me ([email protected]) for more details.

ANALYSIS

Hedge Funds and Prop Shops Converge

This FT story from last week explores how the line between high-frequency trading firms and quantitative hedge funds is blurring. It’s a long read, but worth the time.

High-frequency trading firms, once focused on ultra-short-term strategies, are now extending into multi-day trades to deploy more capital.

Meanwhile, quantitative hedge funds are adopting faster, more systematic trading techniques to boost performance and meet investor demand.

I asked James Corcoran, head of AI at STAC, about this trend:

We see data that confirms the trend toward convergence between HFTs and prop trading firms.

One of the clearest signs is in how STAC benchmarks are being used. Benchmarks for ultra-low-latency trading platforms — once almost exclusively the domain of HFT specialists — are now in demand from prop firms. Meanwhile, our quant trading benchmarks, traditionally popular with the hedge funds, are increasingly accessed by firms you’d classify as HFTs.

This convergence has fuelled demand for the STAC machine learning benchmark, designed to evaluate the performance of platforms generating signals in this converged latency window. There's a lot of action in this space. 

Takeaway

AI could accelerate this convergence by giving both high-frequency traders and quant hedge funds new tools to spot short-lived opportunities.

ADOPTION

JPMorgan’s $2B AI Bet Is Paying for Itself

Jamie Dimon says JPMorgan Chase is now saving about as much as it spends on AI—roughly $2 billion a year.

“We know that it’s got billions of cost savings, and I think it’s the tip of the iceberg,” the bank’s chief executive officer said on Tuesday.

The bank has hundreds of AI use cases across trading, research, hedging, and customer service, with thousands of employees integrating the technology firmwide.

Dimon has likened AI’s importance to the printing press, steam engine, and electricity, and believes it will improve productivity and quality of life, even if it eliminates some jobs along the way.

Takeaway

I'm not sure exactly what kind of ROI timeline JPM is on, but recouping the amount spent seems pretty significant given that ChatGPT rolled out to the broader world fewer than three years ago.

ROUNDUP

What Else I’m Reading

• Why Hedge Funds Are Pouring Back Into Commodities (Rupak Ghose)

• Despite Fears, AI Has Yet to Disrupt the Job Market (Yale Budget Lab)

• CommBank Lures Scammers into Exposing Tactics with AI (PR)

• Goldman CEO: AI Will Create More Jobs, Not Cut Them (Yahoo)

• Bank of America Employs AI to Answer Payments Questions (PYMNTS)

• Morgan Stanley’s Tech Boss Says AI Coding Has ‘Profound’ Impact (BBG)

• OpenAI Valuation Reaches $500 Billion, Topping Musk’s SpaceX (BBG)

• Bud Financial Launches MCP For Bank Data (Fintech Times)

• Ex PayPal, Intuit CEO Harris Launches AI Wealth Platform (Finovate)

• EU to Invest €1 Billion in AI, Cut Reliance on Foreign Tech (WSJ)

• Rogo Appoints Former Lazard MS Rekhi as President (PR)

• AI Drives Record Turnout for WorldQuant Trading Contest (Reuters)

• Snowflake launches Cortex AI for Financial Services, MCP Server (PR)

CALENDAR

Upcoming AI + Finance Conferences

• Open Source in Finance Forum - Oct. 21-22 • New York

  Finance and tech leaders tackle how open source and AI can be governed, scaled, and applied in financial services.

• AIFin Workshop at ECAI 2025 – October 26, 2025 • Bologna, Italy

  One-day academic workshop on AI/ML in finance, covering trading, risk, fraud, NLP, and regulation.

• AI in Finance 2025 – October 27–30, 2025 • Montréal

  Academic event covering ML in empirical asset pricing and risk.

• ACM ICAIF 2025 – November 15–18, 2025 • Singapore

  Top-tier academic/industry conference on AI in finance and trading.

• AI for Finance – November 24–26, 2025 • Paris

  Artefact’s AI for Finance summit, focused on generative AI, future of finance, digital sovereignty, and regulation 

• NeurIPS Workshop: Generative AI in Finance – Dec. 6/7 • San Diego One-day academic workshop at NeurIPS focused on generative AI applications in finance, organized by ML researchers.