After years working on institutional trading, market connectivity and real-time risk technology, Lev Yatsemyrskyi has increasingly focused on another infrastructure challenge facing financial institutions: how artificial intelligence can be deployed at scale without losing control over reliability, permissions, oversight and risk.
There is a side of financial technology that most people never see.
A trading platform may appear straightforward on a screen, but behind it sits an interconnected network carrying orders, executions, market information and risk data between exchanges, brokers, investment firms and internal systems.
When that infrastructure works, it largely disappears from view. When it does not, the consequences can become apparent very quickly.
Much of Lev’s career has been spent in that less visible layer of finance.
During nearly six years at Nasdaq, Lev Yatsemyrskyi worked on institutional trading, connectivity and risk technology supporting major financial-market participants. His responsibilities centered on the Nasdaq Risk Platform and included market connectivity, FIX-based integrations, real-time financial data, institutional onboarding and the production systems supporting those functions.
The terminology can make the work sound more routine than it is.
In many enterprise applications, a software problem can wait for the next release or the following morning. Electronic markets operate differently. If trading information stops flowing correctly, connectivity is interrupted or a downstream risk process is affected, an engineering problem can become an operational one almost immediately.
There may be very little time to determine what went wrong.
That environment shaped Lev’s approach to technology.
Accounts from former colleagues and institutional counterparts familiar with his Nasdaq work describe responsibilities extending across more than 100 institutional client environments on the Nasdaq Risk Platform. Those institutions included hedge funds, market makers, liquidity providers, broker-dealers and other financial firms operating with different technology configurations and requirements.
The number conveys scale, but the more difficult problem sits underneath it.
Institutional clients rarely arrive with identical systems. Connectivity models vary. Internal workflows differ. A firm’s operational and risk requirements may look substantially different from those of another participant, even when both communicate through the same industry protocols.
Scaling technology across that environment requires a balance: enough standardization to support growth, but enough technical understanding to recognize when one institution cannot simply be treated like the one before it.
A Large-Scale Institutional Implementation
One of the larger initiatives associated with Lev’s Nasdaq work involved the OCEAN ATS environment.
Accounts from people familiar with the implementation describe way over 100 integrations into the Nasdaq Risk Platform. The effort was expected to take approximately six months and was completed in slightly under five.
Finishing ahead of schedule matters, although in market infrastructure speed alone is not the difficult part.
The integrations still have to work when clients go live.
Connectivity has to remain reliable. Trading information must arrive where it is expected. Downstream systems need to process that information correctly. When something does fail, engineers must determine quickly whether the source is configuration, connectivity, data processing or another part of the environment.
The significance of work at that scale therefore goes beyond the number of integrations or the implementation timetable. The systems involved were connected to live institutional trading and risk processes, where problems in connectivity or financial-data handling could have immediate operational consequences for participating firms.
And the technical complexity of a problem does not always determine its operational importance.
A complicated issue can sometimes wait.
A seemingly small one may require immediate attention.
Knowing the difference requires more than familiarity with a protocol or software architecture. It requires understanding what the technology is supporting and what happens when financial institutions begin depending on it in a live market environment.
For Lev, that distinction became an important part of how he approached engineering.
From Trading Infrastructure to AI Infrastructure
That experience has increasingly influenced the way Mr. Yatsemyrskyi thinks about artificial intelligence.
Much of the public discussion surrounding AI still focuses on model capability: how effectively a system reasons, generates information, interprets data or performs a particular task.
Lev is interested in what happens after the model becomes capable enough to interact with real systems.
In financial services, the distinction is consequential.
An AI system that produces an incorrect answer creates one type of problem. A system capable of taking an unauthorized action, accessing information beyond its intended scope or interacting with a production environment without sufficient controls creates another.
The difference is that a bad answer remains information until something acts on it. Once an AI system has operational authority, bad reasoning can become a business event.
Lev has explored these questions through his writing on AI infrastructure and governance, including articles published through the Forbes Technology Council.
In “Why AI Governance In Regulated Industries Is An Observability Problem, Not A Policy Problem” he examined the gap between governance policies describing how an AI system should behave and the operational evidence needed to understand what the system actually did.
In “Why The ‘Last Mile’ Of AI In Finance Is An Infrastructure Problem, Not A Model Problem” he focused on the infrastructure between increasingly capable AI models and the financial systems, data and decisions into which those models are deployed.
Together, the articles reflect a recurring theme in Lev’s work: sophisticated models alone are not enough to make AI dependable inside consequential financial environments.
His focus is increasingly on what might be called the last mile of enterprise AI — the infrastructure between an intelligent model and the business systems, information and decisions to which it is eventually connected.
The difficult questions quickly become practical ones.
What information should an AI system be permitted to access?
What authority has it actually been given?
Which actions require human approval?
Can an institution reconstruct what the system did after the fact?
And what happens when its behavior falls outside expected boundaries?
Financial markets have been dealing with variations of these questions for decades.
Long before generative AI entered the enterprise, financial infrastructure had to be built around permissions, monitoring, auditability, operational resilience and risk controls. Systems could not simply be fast or technically sophisticated. Institutions also needed to understand what those systems were doing and maintain boundaries around what they were allowed to do.
It does not eliminate those requirements.
If anything, increasingly autonomous software may make them more important.
This is where Lev sees a connection between the infrastructure developed around electronic markets and the infrastructure institutions will need as AI becomes more deeply embedded in regulated environments.
A capable model is only one component.
What surrounds it may ultimately determine whether institutions can trust it enough to use it for consequential work.
From Practitioner to Industry Conversation
Lev’s work around these issues has increasingly moved beyond engineering environments and into broader industry discussions.
At the QA Financial Forum New York 2026, held at the Harvard Club of New York City, he joined a leadership panel examining “The AI Revolution in Quality Engineering: What Is the Cutting Edge?” alongside technology professionals from Bank of America Merrill Lynch and Morgan Stanley.
The discussion addressed a problem financial institutions are already confronting: what happens when AI moves beyond experimentation and becomes embedded in software development, testing and operational processes.
Questions around agentic AI, risk-based testing, observability and the return on enterprise AI investment may appear distinct, but they converge around the same issue — how institutions determine whether increasingly automated technology is reliable enough to place into production.
Lev later took on a different role at Corinium Global Intelligence’s Chief AI Officer New York 2026, moderating the opening panel, “Trust by Design – Governing AI at Scale in Highly Regulated Enterprises.”
The discussion brought together senior practitioners working across capital markets, insurance, education and other environments to examine how organizations move from general AI principles to governance that can actually be enforced.
For Lev, the subject closely mirrored questions emerging from his own work.
AI governance can be written as policy. Eventually, however, that policy has to become technology.
Someone must translate abstract rules into permissions, monitoring, approval boundaries and audit trails. Accountability has to survive automation, and when something fails, an organization needs enough evidence to reconstruct what happened.
In increasingly autonomous environments, governance may also have to operate before the action occurs — defining whether a system has authority to perform an action in the first place rather than relying entirely on reconstruction afterward.
That is where governance begins to look less like documentation and more like infrastructure.
The speaking and moderation roles reflect a broader shift in Lev’s work — from developing systems internally to participating in industry discussions about how those systems should be governed and deployed.
Evaluating the Work of Other Technology Builders
His industry work has also expanded into evaluating emerging technology companies.
On September 10, 2026, Lev Yatsemyrskyi served as an Assessor and Judge for the Seongnam Korean New York Market Entry Deeptech Accelerator Program organized by Elpis Labs in New York City, where he evaluated 10 South Korean startups.
The assessment was conducted using a structured scorecard covering technical and commercial criteria including innovation, engineering depth, product execution, scalability, business viability, regulatory readiness, safety considerations and investment potential. Lev completed the evaluation process by assigning numerical scores and qualitative assessments to the participating companies.
It is a different role, but the questions are familiar.
A startup may have an impressive technical demonstration.
That alone does not make it a viable product.
Can the system function reliably outside the demonstration environment?
Can it scale?
Does the team understand how customers will actually use it?
And when technology enters finance, healthcare or another regulated environment, have the founders considered the constraints that appear once the product encounters the real world?
Those distinctions become increasingly important as advanced technologies move from research and prototypes into commercial systems.
When Software Begins to Act
The same issue becomes more complicated as AI moves from generating information toward executing tasks.
Traditional enterprise software generally operates through workflows designed explicitly in advance. Agentic AI introduces a different model.
Software may interpret an objective, determine intermediate steps, interact with multiple systems and decide how to proceed.
That creates significant opportunities for automation.
It also creates a control problem.
For a financial institution, the important question is no longer simply whether an AI system can perform a task. The institution must also determine whether it can define and enforce the boundaries within which that task is performed.
Lev has argued that authority therefore needs to become part of system architecture itself.
A policy stating that an AI system should not perform a certain action offers limited protection if the technical infrastructure still gives that system the ability to perform it.
The distinction has led him to focus on observability, authorization, human oversight and the ability to reconstruct AI-driven activity after it occurs.
But observability and authorization solve different problems.
Observability helps an institution understand what happened.
Authorization determines what the system should have been permitted to do in the first place.
As AI systems acquire greater operational authority, institutions may need both.
The thinking reflects a broader theme across Lev’s career.
The more consequential a technology becomes, the less its reliability and control mechanisms can be treated as secondary considerations.
Finance’s Next Infrastructure Problem
After Nasdaq, Lev Yatsemyrskyi continued his work in quantitative finance, ultimately moving into a role as Director of Quantitative Technology at Qube Research & Technologies (QRT).
The setting changed, but many of the underlying questions remained familiar: real-time information, automation, system reliability and the relationship between increasingly sophisticated technology and financial decision-making.
Those questions are becoming more important as AI moves deeper into financial institutions.
The larger issue is no longer whether financial firms will use artificial intelligence.
They already are.
The more difficult question is what kind of infrastructure will be required as those systems become connected to increasingly important decisions and operational processes.
Financial markets offer a useful precedent.
Electronic trading did not reach institutional scale simply because algorithms became more sophisticated.
An extensive layer of infrastructure developed around them: connectivity standards, risk controls, monitoring systems, permissions, audit trails and operating procedures for situations in which something did not behave as expected.
That infrastructure made sophisticated automation usable at institutional scale.
AI may now require an equivalent layer.
For Lev, some of the most consequential work in financial technology is likely to happen there – between what an intelligent system is technically capable of doing and what an institution is actually prepared to allow it to do.
The models will continue to attract most of the attention.
But financial technology has taught the same lesson repeatedly.
The systems attracting the least attention are often the ones everything else depends on.
