From Cloud Modernization to Intelligent Infrastructure: How Shubh Prabhat Is Connecting Enterprise Engineering with AeglysAI

For much of the last two decades, enterprise technology modernization has been defined by a familiar progression: monolithic applications gave way to services, data centers expanded into cloud platforms, and manually operated infrastructure became increasingly automated. Today, artificial intelligence is introducing another transition—one that may require engineers to rethink not only how systems are built, but how they make decisions.

Technology architect and researcher Shubh Prabhat has spent approximately 18 years working across this evolution. His experience spans software engineering, application and solution architecture, cloud-native platforms, microservices, distributed systems, identity and authorization, cybersecurity, network automation, and AI-enabled engineering.

Those areas now converge in his work on AeglysAI, an open-source research and engineering initiative exploring adaptive intelligence for secure and resilient distributed systems.

Engineering Through Multiple Technology Cycles

Prabhat’s career has involved enterprise technology environments across sectors including telecommunications and financial services. A recurring challenge throughout that work has been modernization: how to transform complex systems while maintaining reliability, security, and continuity.

In one large-scale modernization initiative, Prabhat contributed to the migration of microservices with zero downtime, while the resulting deployment approach reduced deployment effort , according to project metrics maintained from the initiative.

Experiences like these shaped a broader engineering perspective. Modernization, in this view, is not simply about replacing older technology with newer frameworks. The harder challenge is preserving predictable system behavior while architecture, infrastructure, and operational processes change around it.

That problem becomes even more significant as AI enters infrastructure and security decision-making.

Moving From Automated to Adaptive Systems

Traditional automation is largely deterministic. Engineers define conditions, and software executes predetermined actions when those conditions occur.

AI introduces the possibility of systems that can identify patterns, evaluate changing conditions, and produce risk or operational intelligence dynamically.

But this creates a difficult architectural question: How much decision-making authority should intelligent models receive?

For security-critical distributed systems, Prabhat’s research direction argues for a separation between intelligence and authority.

An AI or behavioral model might determine that an access request exhibits unusual characteristics. It could identify an unfamiliar device, abnormal request frequency, unusual access time, or deviation from historical behavior. But the model’s assessment does not necessarily need to become the final security decision.

Instead, the intelligence can become evidence evaluated within deterministic policy boundaries.

This principle—AI advises, while policy decides—has become central to the research direction behind AeglysAI.

Connecting Enterprise Engineering With AeglysAI

AeglysAI is being developed as an open-source environment for investigating how intelligent capabilities can participate in distributed-system control loops without replacing deterministic engineering safeguards.

Its research direction brings together several areas that are often treated separately: authorization, behavioral risk, operational context, observability, resilient systems, and AI-assisted decision-making.

One area of investigation is the progression from traditional role-based access control toward increasingly contextual authorization.

Role-Based Access Control, or RBAC, remains effective for determining what actions a particular role is permitted to perform. Attribute-Based Access Control adds additional information about users, resources, and environments.

Adaptive systems introduce another dimension: changing risk.

A request that appears acceptable based on identity and role alone may look very different when behavioral and operational context is considered.

The research challenge is therefore not simply to produce another risk score. It is to determine how that intelligence can safely influence authorization and response while keeping decisions explainable, auditable, and bounded by policy.

Building for Explainability and Resilience

For Prabhat, explainability is an engineering requirement rather than an optional AI feature.

If a distributed security system changes an authorization decision because risk increased, engineers should be able to understand why.

Was the device unfamiliar? Did the request frequency change? Was a sensitive resource being accessed outside the user’s normal behavioral pattern? Which policy converted those signals into an allow, step-up, limit, or deny decision?

Capturing this reasoning can improve debugging, auditing, governance, and incident investigation.

Resilience introduces another question: what happens when the intelligent component itself becomes unavailable?

A security architecture should not become dependent on an AI model to the point that model failure creates uncontrolled behavior. Policy-bounded design therefore also requires deterministic fallback mechanisms and clearly defined failure behavior.

These principles reflect lessons from enterprise engineering, where reliability frequently depends as much on how systems fail as on how they operate under normal conditions.

Additional implementation details, architectural specifications, experimental resources, and supporting technical documentation can be found on the AeglysAI website and in the associated public source-code repository.

From Research Platform to Open Engineering

AeglysAI is also intended to make this work reproducible and open to technical scrutiny. The initiative is developing reference implementations, architectural documentation, experiments, and an engineering community around adaptive security and resilient distributed systems.

Its planned community hackathon series extends that idea by inviting students, researchers, and engineers to explore problems involving adaptive authorization, behavioral risk, and resilient response. The initiative is structured around the broader engineering cycle “Observe. Assess. Authorize. Respond.”

For Prabhat, this represents a continuation of the same problem that has shaped much of his enterprise work: systems must evolve without losing the controls that make them dependable.

The difference is that the next generation of modernization may not simply involve moving applications to new infrastructure. It may involve building infrastructure capable of interpreting changing conditions intelligently—while ensuring that intelligence remains explainable, measurable, and bounded by engineering policy.

As AI becomes more deeply embedded in enterprise systems, that distinction could become increasingly important.