AI in Hedge Funds: What AI Engineers Build and How the Funds Hire Them

An AI engineer at a hedge fund builds and runs the machine-learning systems that researchers and portfolio managers work with: the training libraries, the data pipelines, the research tools built on large language models and, more and more, agents that carry out research tasks on their own. The big quantitative and multi-strategy funds hire for the seat, and the interview processes the firms publish are built around code, not finance.

What those systems can now do is the surprising part. In June 2026, Ken Griffin, Citadel's founder and chief executive, told a Goldman Sachs audience that a colleague had "built an agentic AI system that would read a paper, reproduce it, verify the results that were published in the paper, produce the results out of sample, and do all of this work in about, on average, two to three hours per paper."

Reproducing a paper the usual way, he said, "takes roughly six to eight weeks." And "there's no reduction to headcount at Citadel on the back of this breakthrough."

What does an AI engineer at a hedge fund do?

The job is building the things other people use to research and trade. Jane Street, which calls itself "a research-driven trading firm," gives the plainest account of the room it happens in: "No silos. Researchers, engineers, and traders sit a few feet away from each other and work together to train models, build systems, and run trading strategies."

The work, day to day

On a given day, in Jane Street's words, the team might "dive deep into market data, tune hyperparameters, debug distributed training, or study our model's trades in production." The firm builds on "the latest papers in LLMs, computer vision, RL, training libraries, cuda kernels, or whatever else we need to train good models," and invents its own architectures and optimizations for trading.

The difficulty is the data. Its research internship describes "petabytes of data, produced by adversarial markets, that evolve everyday. Signals are small, noise is high."

Tens of thousands

High-end GPUs

Jane Street

1 exabyte+

Data storage

Jane Street

900K+

Data files ingested a day

Millennium

1,800+

Technologists

Millennium

Jane Street also splits machine learning into three jobs, and the split is the best short map of the seat:

• ML researchers "invent and reinvent the methods and models driving our trading strategies."
• ML research engineers "bring our models to life with the best tools for the job."
• ML performance engineers "speed up our models, automate them, and manage the care and feeding of our training loops."

The AI engineer's seat is the second and third of those. The dividing line with a quant researcher is ownership: the researcher owns the trading idea, the engineer owns the code, the infrastructure and the speed it runs on, and the two sit close enough to argue about it across a desk.

One project, start to finish

Here is an illustrative first project, assembled from the duties the firms' postings list rather than taken from any one firm: turning a stream of company filings into data a researcher can test.

1. Pull the documents in and store each one with the moment it became public, so no model ever sees a filing before the market did.
2. Use a language model to extract one field from each filing, say a company's guidance range.
3. Grade the extraction against a hand-labelled set before anyone relies on it. Say the model gets 460 of 500 labelled filings right: that is 92% accuracy, and the 40 misses are where the week goes, working out which formats break it and whether the errors are random or lean one way.
4. Hand the researcher a clean table, then make the job fast and cheap enough to run every day.

The researcher decides whether guidance ranges predict anything. The engineer makes sure the answer is not an artefact of bad data.

What the first year looks like

The internships are the closest thing to a published first year. Jane Street's machine learning engineering interns are paired with full-time mentors on "real-world ML projects we actually need done," with access to a GPU cluster of "thousands of H100/H200/B200s." The stated lesson is the gap between textbook machine learning and what happens when it meets noisy financial data.

The bar is practical rather than academic: an undergraduate or PhD student "with practical experience training an ML model, working on an ML library, or optimizing an ML workflow." For students heading into research, the same firm funds a Graduate Research Fellowship: a year of tuition and fees plus a $50,000 stipend.

Who hires AI engineers at hedge funds?

The seat exists across the big quantitative and multi-strategy funds, and at the trading firms beside them, such as Jane Street and Citadel Securities. What differs is where it sits: beside the researchers, beside the portfolio managers, or in a lab of its own.

Beside the researchers

A 2026 Citadel Securities posting for a machine learning engineer described a job that is entirely a tool for other people: "the next generation library for deep learning within the firm," built with researchers and with high-performance computing specialists to cut training time and cost. The work, it said, "will empower 100+ researchers to iterate faster." It asked for two years of machine learning and software experience.

Two Sigma, where more than half the staff are engineers, has a job family built on the same idea: modeling engineers "build the platforms and tools that our modelers use to ideate, create, test, and deploy our quantitative strategies." Its generative AI engineers sit in a News Engineering team whose aim is to "collect, organize, and monetize the world's unstructured data, including news, social media, and regulatory filings."

Beside the portfolio managers

Other seats serve investors directly. Citadel built the Citadel AI Assistant, a chatbot trained on licensed outside content such as transcripts, filings and brokerage research, and on the hedge fund's own investment strategies. Chief technology officer Umesh Subramanian told Reuters in December 2025 that nearly all of the firm's equities investors used it regularly, with investment judgment still in human hands: "AI is a tool investors are going to use, and how you use it will drive performance."

Balyasny runs the same idea as a hub and spokes. Its Applied AI team is, in the firm's words, "a group of researchers, engineers, and domain experts who are building AI-native tools that embed directly into team-level workflows."

OpenAI's case study on the firm, published in March 2026, fills in the structure. The team was set up in late 2022; it builds the shared frameworks, toolchains and compliance guardrails, and each investment team builds its own tools for its asset class on top, an arrangement the case study calls "federated deployment."

Man Group puts the AI researcher inside an investment team. A 2026 Man Group posting placed its quantitative researcher for AI and machine learning in its China research team in Shanghai, using large language models, natural language processing and alternative data to find new trading signals.

In a lab of its own

Bridgewater went furthest. AIA Labs, spearheaded by co-chief investment officer Greg Jensen, calls itself the firm's "dedicated Artificial Intelligence research and investment lab," and it is building "an artificial investor that aims to meet or exceed expert human performance across the full range of activities that investors perform."

The lab works "in deep partnership" with Pure Alpha, Bridgewater's flagship strategy, and the effort is older than the chatbot boom: Bridgewater hired Jasjeet Sekhon, from Yale, as chief scientist for it in 2018.

How AI is changing the hedge fund engineer's job

Bridgewater's artificial investor is the far end of a wider shift: the systems engineers build are starting to do research tasks themselves.

From models to agents

Man Group's own account of that system, AlphaGPT, is "a digital three-person research team that never sleeps." One member proposes hypotheses, one turns them into code and one judges the results against strict criteria, all run by a workflow orchestrator with safeguards against "AI's common pitfalls such as hallucinations."

Balyasny's tools take on narrower jobs. OpenAI's case study credits one, a Central Bank Speech Analyst, with cutting macroeconomic scenario analysis from two days to about 30 minutes. The paper-reproducing agent in the opening belongs to the same family.

Part of the job is spreading the tools. In 2026, Balyasny's Applied AI team ran a hackathon open to all staff "regardless of previous coding experience," in which participants worked with the team to automate complex tasks. The winners included teams from investing, data science and office services.

For the engineer, the work moves up a level: less building one model, more building the orchestration, evaluation and guardrails that let a chain of models do a researcher's first pass without going wrong.

Who signs off

Human review has not gone away, even where the machine leads. At Man Group, every AI-produced signal must "demonstrate clear economic rationale and pass identical evaluation thresholds before it can be considered for deployment."

Bridgewater's lab frames its output as "tools that empower human investors," and its forecasting research points the same way. The AIA Forecaster matched human superforecasters on one public benchmark and trailed market consensus on a prediction-market benchmark, but an ensemble of the two beat consensus alone.

The analyst's version of the same question, whether AI replaces the people doing the work, is taken up in Will AI replace investment banking analysts.

What skills does a hedge fund AI engineer need?

The skills follow from that work: code that runs fast, models that train at scale and, more and more, agents that can be trusted with a research task. The firms' 2026 postings name this toolkit.

The machine-learning core

Research-heavy seats go deeper into theory. Man Group's 2026 posting for an AI and machine learning researcher asked for a "background in core ML theory: optimization, probabilistic modeling, deep learning architectures (transformers, diffusion models, etc.)" and hands-on experience fine-tuning large foundation models.

Engineering seats weight the systems side instead: distributed training, GPU performance, and the shared libraries that let a researcher's experiments run faster and cheaper.

Agents and evaluation

Then there are agents. D. E. Shaw's 2026 applied AI engineer posting asked for "experience with AI agents and agentic frameworks, including orchestration, tool use, memory, and multi-step reasoning in production environments." The same engineer would build AI coding agents and developer tools to speed up the firm's own software work, plus shared pieces such as retrieval systems.

Evaluation is part of the same job. The posting's first duty begins "Build, evaluate, and deploy," and Man Group's AlphaGPT gives one member of its three-person team to evaluation alone.

How do you become an AI engineer at a hedge fund?

Those skills arrive by one of two routes, and the postings describe both: straight from a degree, often a doctorate for the research seats, or across from software and machine-learning work somewhere else.

What the postings ask for

The engineering seats are relaxed about degrees and strict about experience. D. E. Shaw's applied AI engineer posting asked for a bachelor's degree "in any field, along with an extensive background in software development and hands-on experience building and scaling AI solutions." Point72's generative AI engineer posting asked for three to seven years of AI and machine-learning engineering; Two Sigma's generative AI engineer needed at least one year, with three to ten preferred.

Research seats set a different bar. Two Sigma's campus AI research scientist role preferred a PhD, or a master's with some work experience, and listed "publications at NeurIPS, ICML, ICLR or similar" as preferred.

Campus routes

For students, the front door is an internship or a graduate programme. The machine learning internships described above are the most detailed published version of the route.

Millennium runs LEaD (Learning Engineering and Data), a 12- to 18-month rotational programme for early-career software engineers in Miami, and its own careers portal listed a 2027 Applied AI Engineer internship there. Two Sigma folds its campus hires in engineering, data science and quantitative research into one programme, NewTS, and in 2026 advertised an AI research scientist role for campus hires.

Lateral routes

The other route runs across from AI labs and other engineering work. Bridgewater's lab is explicit about where its people come from: "We come from frontier AI labs, academia, and the investment community," it says, and it wants "people dedicated to elegant engineering, creative research." For an engineer with years of shipped machine-learning systems, the experience-based postings above are the door.

What the postings pay

Several firms print a base salary range on the posting itself. In September 2026, the firms' own postings listed these:

• Two Sigma, generative AI engineer: $165,000 to $300,000
• Two Sigma, AI research scientist hired from campus: $300,000
• D. E. Shaw, applied AI engineer: $225,000 to $275,000
• D. E. Shaw, machine learning researcher: $275,000 to $350,000
• Point72, generative AI machine learning engineer: $185,000 to $300,000

What do hedge fund AI engineer interviews test?

Whichever route you come by, the interview is built around code. Every process these firms publish tests it, and most open with it; machine learning, where it appears, comes as an added round on a software loop.

Coding comes first

Jane Street is blunt about its software loop: "We don't ask software engineers to do mental math, or math olympiad questions, or to contemplate logic puzzles. SWE interviews are about programming, plain and simple." It expects real code, not pseudocode. Two Sigma candidates code in C, C++, Java or Python alongside the interviewer, and Citadel's quantitative research loop names Python and C++ as its core languages, "but we welcome any languages."

Where the machine learning comes in

Jane Street's machine learning engineering internship is the clearest published case. It runs the software intern loop "with one key addition": after the Zoom coding interview, an onsite day with rounds dedicated to assessing machine learning engineering skills.

Finance is optional; reasoning is not

None of these published processes lists finance as a topic, and Jane Street tells machine learning interns it does not expect a finance background. What the firms ask for instead is visible reasoning. Citadel's advice to candidates: "Share your thought process. Discuss trade-offs among different possible approaches. Be clear about what you know and what you don't know."

Two Sigma says much the same to research candidates: "Don't be afraid to give a 'first pass' solution and then iterate." Its recruiter prep call is the place to ask exactly which topics the loop will cover.

How to prepare for an AI engineer role at a hedge fund

The preparation that works is concrete: things you have built and can explain line by line.

1. Ship one model end to end. Train it, serve it, monitor it, and keep notes on what broke. That is the practical experience the student postings describe, and it gives an interviewer something real to probe.
2. Make a training run faster. Profile a job you already have, find the bottleneck in data loading or GPU use, and measure the gain. Performance work is a job family of its own at some firms.
3. Build a multi-step LLM tool and grade it. Give it one research task, write the evaluation before you tune it, and record where it fails. That is the shape of the work the applied AI postings describe.
4. Test on noisy, out-of-sample data. Backtest one idea on market data and report where it stops working. A model that fails honestly teaches more than a clean benchmark score.
5. Read what the firm has published about its own AI work. A lab page, an engineering blog or a technology chief's interview gives you a specific answer to "why here?"

A structured self-check lives on the interview readiness tool. For the bank side of the same shift, the page on JPMorgan's LLM Suite covers what the bank built and the engineering skills it hires for, and the wider library of guides covers the rest.

The bottom line

An AI engineer at a hedge fund is paid to build what research runs on: the libraries, the pipelines, the assistants and, increasingly, chains of models that do a researcher's first pass. The firms hire for it the way they hire software engineers, on code and reasoning rather than finance, with doctorates mattering mainly for the research seats.

The paper-reproducing agent from the opening shows where the seat is heading. It did in an afternoon what takes weeks by hand, and the firm kept its researchers. In Griffin's words, "with the talented people we have, we just have more to go after."