Balyasny builds AI research platform for hedge fund teams

Balyasny Asset Management has built an AI-driven investment research platform that now sits at the centre of day-to-day workflows for most of its investment teams, as hedge funds and asset managers try to keep up with rising volumes of market and corporate data.

The multi-strategy firm runs about 180 investment teams across asset classes and geographies. It has positioned the system as a way to make research faster and more structured, while operating under the tight controls required in regulated markets.

A 20-person Applied AI group, formed in late 2022, led the build. The team includes researchers, engineers and domain experts. It designs shared tools and guardrails while allowing individual desks to tailor agents to their strategies.

Charlie Flanagan, Balyasny's Chief AI Officer, described the impact on research workflows:

"AI is enabling our teams to apply first principles thinking faster, across more data, and with more structure," said Flanagan.

Research pressure

Investment research typically requires analysts to work across large collections of material, including market data, broker research, regulatory filings, earnings materials and records of expert calls. Many firms have experimented with general-purpose AI tools, but often struggle to combine structured and unstructured sources, control workflows, and meet internal compliance requirements.

Balyasny has framed its effort as purpose-built for institutional use. It says the system can reason across sources, retrieve material and take actions through tools in a way that resembles a skilled analyst's workflow.

Model testing

A key part of the programme has been formal model evaluation before deployment. Balyasny built an internal evaluation pipeline that measures models across more than 12 dimensions, including forecasting accuracy, numerical reasoning, scenario analysis and robustness to noisy inputs. The process uses internal benchmarks and proprietary financial data.

The firm uses OpenAI models as part of the stack. It says testing highlighted strengths in the GPT-5.4 family for multi-step planning and tool execution, and showed lower hallucination rates than other options. Balyasny now uses GPT-5.4 as a reasoning engine alongside internal models, selecting models by task performance.

Su Wang, a senior research scientist at Balyasny, tied the choice to the testing framework.

"We evaluate models the way we evaluate investments: on fundamentals. GPT-5.4 proved it could plan, reason, and execute with real rigor," said Wang.

Workflow design

Balyasny also worked with OpenAI on how models behave in real analyst workflows. The firm said OpenAI teams observed investment teams using the internal platform and used that input to guide iterations.

Jonathan Park, a product manager at Balyasny, said the approach improved the feedback process.

"We didn't just tell OpenAI what we needed. We showed them. And that made all the difference," said Park.

The firm also emphasised ongoing internal feedback loops. It collects structured feedback on outputs, user assessments and tool execution, and runs outcome audits. Those signals are used to adjust models and orchestration rather than treat the platform as a static product.

One example came from merger arbitrage. The desk needed agents that could update assessments as new filings and press releases arrived. Balyasny said it expanded agent planning and tool access so the workflow could move from manual spreadsheet tracking and alerts to continuous monitoring of deal probabilities.

Central build

Balyasny's architecture combines central development with local customisation. The Applied AI team maintains core components such as agent frameworks, shared toolchains and compliance guardrails, then deploys them to investment teams with controlled access to datasets and tools.

The firm describes this as a federated deployment model. Each team can configure agents for its asset class, such as macro, commodities or equities, while central guardrails address firm-wide requirements for risk controls, data security and regulatory compliance.

Kevin Byrne, Balyasny's Chief Operating Officer, linked the structure to adoption across desks.

"Our early investments in AI paid off. Today, every one of our investment teams can decide how to apply the latest AI to their process, in a secure environment and with real-time expert guidance," said Byrne.

Use cases

About 95% of Balyasny's investment teams now actively use the platform, according to the firm. It reports time savings on research tasks that previously took days, and says agents can synthesise tens of thousands of documents, including filings, broker research, earnings materials and expert call notes.

In one example, Balyasny said a "Central Bank Speech Analyst" reduced the time spent on macroeconomic scenario analysis from two days to about 30 minutes. Another agent, described as a "Merger Arbitrage Superforecaster", continuously monitors and updates deal probabilities.

The firm also pointed to analyst confidence as a benefit. It says the system uses scoped tools, traceable reasoning paths and testable agents-features designed to produce structured, explainable outputs that feed into human decision-making.

Charlie Sweat, a portfolio manager at Balyasny, compared the system to a new colleague embedded in the workflow.

"It's like adding a teammate who never forgets, always cites sources, and double-checks the details before sending anything back," said Sweat.

Next steps on the roadmap include reinforcement fine-tuning, expanded agent orchestration across financial domains, multimodal inputs such as charts and filings, and continued evaluation of newer frontier models for fit with investment research tasks.