We are thrilled to welcome Michael to the team as Transient’s new CTO.
I sat down with Michael to discuss a journey spanning Wall Street trading systems, hyperscale cloud infrastructure, AI, and logistics at Amazon, and why the next challenge in financial technology isn’t simply building smarter AI, but creating the infrastructure that allows increasingly capable AI systems to operate safely inside financial institutions.
Jennifer: Welcome aboard, Michael! To kick things off, tell us a bit about your journey before joining Transient.AI.
Michael: Thanks! My technical journey began in 2003 when I moved from Pondicherry, India, to the U.S. to pursue my Master’s degree in Computer Science.
I started my career as an intern at Citigroup, focusing on fixed income trading desk technology and building electronic trading platforms. I later joined Credit Suisse, where I spent several years building risk management systems, ultra-low-latency infrastructure, and request-for-quote (RFQ) systems connecting brokers and dealers.
I then relocated to Seattle to join Amazon, where I spent nearly a decade. During my time there, I worked across their AWS cloud division building new cloud services, engineered AI features for their voice assistant, Alexa, and led engineering for last-mile delivery logistics, building the mobile delivery application and backend orchestration that powers delivery operations.
Jennifer: After nearly 10 years at Amazon, what drew you to Transient?
Michael: I chose Transient.AI because I believe the next phase of AI in finance isn’t primarily a model problem; it’s a systems and control problem.
Financial institutions are operating some of the world’s most complex technology environments, where decisions can have immediate financial, operational, and regulatory consequences. These environments were built around explicit permissions, deterministic execution, predictable latency, and complete auditability.
AI introduces something fundamentally different: probabilistic reasoning. The challenge is figuring out how to allow increasingly capable AI systems to operate within infrastructure that was designed to be deterministic.
Wall Street isn’t new to AI or machine learning. Financial institutions have used quantitative models and machine learning techniques for decades across trading, risk management, fraud detection, and other complex workflows. What’s different today is that generative AI and agentic systems have dramatically expanded what software can reason about and potentially act upon.
The technology has been democratized. The hard part is no longer simply accessing a powerful model. The hard part is giving that model the right context, the right tools, the right permissions, and the right controls to operate safely inside a complex institution.
That requires deep understanding of financial workflows, risk, latency, compute economics, governance, security, and auditability. Generic AI systems can provide intelligence, but intelligence alone isn’t enough when the system is operating against production financial infrastructure.
Hyperscalers provide increasingly powerful models and foundational security primitives. The remaining challenge is applying those capabilities to the domain-specific decisions, workflows, and actions that AI agents perform inside financial institutions.
That’s where I believe Transient can play a critical role: building the control layer/Operating system that allows financial institutions to adopt increasingly capable AI without giving up the governance, security, and operational controls they require.
Jennifer: How does Transient’s platform solve those compliance and security concerns for institutions?
Michael: I think about Transient as a runtime control layer for AI.
The traditional view of AI safety is a guardrail: detect something dangerous and stop it. That’s necessary, but it’s not sufficient when AI agents begin interacting directly with enterprise systems.
An agent needs to operate within a controlled loop:
Identity → Context → Policy → Reasoning → Tools → Verification → Action → Audit
The platform needs to understand who the agent is, what information it is allowed to access, which tools it can invoke, what policies apply to the action it is proposing, and whether the resulting action is safe to execute.
In other words, we don’t want to blindly trust the model’s output. We want the system around the model to continuously evaluate what the model is trying to do.
That’s where the circuit-breaker concept becomes important. If an agent attempts an action outside its authorization, violates a policy, produces an unexpected result, or encounters a condition that requires human judgment, the platform can stop or escalate the action before it reaches a production system.
Data sovereignty is another critical part of this architecture. Our platform runs directly inside the client’s own cloud network boundary, whether AWS, Azure, or GCP, so sensitive enterprise data and agent execution remain within the institution’s controlled environment.
But I think the industry needs to think beyond data sovereignty. We also need execution sovereignty: knowing exactly what an AI system is allowed to do, where it can act, and under what conditions it can take action.
We use controls to protect sensitive information such as personally identifiable information (PII) and proprietary trading data, encrypt data throughout the system, and explicitly configure models so that client data is not stored or used to train external models.
Every consequential action also needs to be explainable and auditable after the fact—not just from the perspective of what the model said, but what context it received, what policies were applied, which tools it used, and what action ultimately occurred.
Jennifer: What’s your focus right now as CTO?
Michael: Right now, customer adoption and collaborative feedback are our top priorities. We work closely with our forward deployment engineers who integrate our platform with clients’ real-world environments.
That creates an important feedback loop between the technology we’re building and the problems financial institutions are actually trying to solve.
We’re particularly focused on understanding where agents can move beyond answering questions and begin safely executing real workflows. That transition from AI that provides information to AI that takes action is where the engineering and governance requirements become significantly more demanding.
Our goal is to continuously improve the runtime, expand the policy and verification capabilities around agents, and make it easier for financial institutions to deploy increasingly capable AI while maintaining control over how those systems operate.
I believe the winning AI infrastructure for finance won’t be defined solely by which model it uses. It will be defined by how safely, reliably, and efficiently that intelligence can operate within the institution.
Jennifer: When you’re not engineering AI infrastructure, what keeps you grounded?
Michael: Family is a big one for me. My wife and I live in Austin, Texas, along with our kids and our dog, Toby. Whenever I need to reset or clear my head after a busy day of building, I love heading out for a run
