Rajan Kohli, CEO of CitiusTech – Interview Series: A Return Conversation

Rajan Kohli is the Chief Executive Officer of CitiusTech and is responsible for the company’s global strategy, growth, and mission of accelerating innovation across healthcare and life sciences. A veteran technology executive with more than three decades of experience, Rajan has led large-scale digital transformation initiatives spanning healthcare, engineering, cloud modernization, data platforms, and artificial intelligence. Since taking the helm at CitiusTech, he has focused on helping healthcare organizations move beyond digital transformation toward intelligence-driven operating models powered by AI, interoperability, and advanced analytics.

CitiusTech is a leading provider of healthcare technology services, consulting, and digital solutions serving healthcare providers, payers, MedTech companies, and life sciences organizations worldwide. The company specializes in healthcare data platforms, interoperability, cloud modernization, digital engineering, analytics, and artificial intelligence. As healthcare organizations increasingly seek to operationalize AI at scale, CitiusTech has expanded its focus toward building intelligence-driven healthcare systems that combine trusted data foundations, governance, interoperability, and workflow-aware AI. Through solutions such as Knewron and its broader AI and data ecosystem, CitiusTech helps organizations transform fragmented healthcare operations into connected, context-aware environments that improve efficiency, decision-making, and patient outcomes.

This interview serves as a follow-up to our previous conversation with Rajan Kohli, where we explored the growing role of generative AI, interoperability, healthcare data modernization, and digital transformation across the healthcare ecosystem. Since then, the industry has rapidly progressed from experimenting with AI to deploying it within production environments, creating new challenges around governance, trust, explainability, and operational scale. In this latest discussion, Rajan shares how healthcare organizations can move beyond isolated AI pilots toward intelligence-driven care, why context engineering is emerging as a critical foundation for healthcare AI, and what it will take to build trusted, scalable systems capable of supporting the next generation of patient care.

You bring decades of experience leading large‑scale digital transformation initiatives. How has that journey shaped your perspective on why healthcare transformation is uniquely complex?

Healthcare transformation is uniquely difficult because there is no single definition of success. Clinical outcomes, accurate billing and payments, access, cost and experience often pull in different directions. Unlike other sectors, healthcare is constrained by clinical risk, regulatory scrutiny and ethical accountability. Failure modes are measured in patient outcomes, not revenue leakage.

This is where context engineering becomes foundational: the discipline of structuring the information environment in which AI operates, so outputs are clinically accurate, workflow-aware and compliance-ready from the ground up. Scaling requires systems thinking across value chains rather than digitizing siloed functions. Healthcare requires deep contextualization of workflows, data semantics across clinical, claims, and device data, as well as complex compliance pathways. Platforms like Knewron are built on exactly this principle, moving beyond general-purpose AI to embed domain-specific context at the architecture level.

This complexity is amplified by fragmented ecosystems involving payers, providers, MedTech, and life sciences. Each of these players operates on different systems, incentives, and data standards. What appears to be a technology problem on the surface is almost always a context and coherence problem underneath.

Ultimately, transformation success depends not just on technology modernization but on trust engineering. We must actively engineer a trust layer into global healthcare AI, designing systems that clinicians and regulators can implicitly rely on in real-world care settings. Context engineering is the mechanism that makes this trust possible. When AI understands the full clinical and operational context of a decision, it earns that trust systematically, not by chance.

You’ve described healthcare as reaching an inflection point. What specific forces are driving this shift right now?

The inflection point we are experiencing right now is driven by operational stress, not technological novelty. Cost pressures, clinician burnout, and severe workforce shortages are forcing health systems to rethink operating models entirely instead of just digitizing existing ones. Manual processes no longer scale, particularly in claims, billing, and clinical operations, which is why AI is being aggressively pulled into production environments.

But pulling AI into production is only half the equation; the other half is ensuring AI operates with the right context. Without context engineering, AI in clinical and operational workflows risks generating outputs that are technically correct but clinically or administratively misaligned.

Concurrently, regulatory forces like CMS interoperability rules, price transparency, and digital quality measures are mandating data liquidity and accelerating modernization. Cloud adoption has crossed a maturity threshold, making enterprise-scale modernization and compliance-ready AI deployment feasible. AI is now fully aligned to healthcare readiness because we finally have digitized workflows, richer data assets, and clearer accountability frameworks than in previous technology waves.

What has been missing until now is the connective tissue between raw data assets and meaningful AI action, and that connective tissue is context. We are seeing a distinct market demand for solutions that are purpose-built to help healthcare organizations turn this operational stress into measurable efficiency.

Healthcare has long lagged in digitization compared to other industries. What has changed to make AI deployment at scale possible today?

AI now scales because healthcare organizations are learning to codify policy, clinical guidelines, and operational logic, rather than ingest vast amounts of data. The fundamental shift is moving from models that interpret raw information to systems that execute programmable knowledge with strict oversight. This is precisely what context engineering enables, moving AI from passive interpretation to active, governed execution by structuring the knowledge environment in which models operate.

Foundational friction has been drastically reduced by the widespread adoption of FHIR, HL7, cloud-native data platforms, and event-driven architectures. Healthcare organizations increasingly recognize that AI must be embedded within workflows rather than exist as standalone dashboards. Workflow-aware AI is key to enabling this, ensuring systems understand workflow semantics and drive actionable outcomes.

Tools like MLOps, DevSecOps, and compliance automation have enabled continuous validation, monitoring, and controlled retraining. We are seeing a distinct shift from experimentation to value-linked use cases such as care gaps, prior authorization, claims integrity, imaging, and clinical decision support. This scale is only possible when we embed robust guardrails and human-in-the-loop oversight directly into these operational workflows.

The industry is adding jobs month on month, and that helps because AI is not seen as taking jobs away. Scaled AI is helping direct the investments for better outcomes for patients and clinicians.

Many organizations remain stuck in pilot phases. What are the key barriers preventing AI from moving into real operational use in healthcare?

AI struggles to scale in healthcare because multiple structural realities surface simultaneously. Healthcare processes rarely operate toward a single objective; clinical outcomes, cost, access, patient experience, reimbursement accuracy, and long-term risk often compete, involving providers, payers, regulators, and patients. Defining success is complex but essential. Many pilots fail because they are not anchored to clear, shared outcome metrics. Structured operational context helps address this by codifying goals, constraints, and stakeholder priorities upfront rather than relying on models to infer them from raw data.

A second barrier is the cost and effort of annotation and validation. Scaling requires continuous involvement from clinicians and revenue cycle experts, whose time is both limited and expensive and often underestimated during pilot phases.

Third, governance and integration gaps become more visible at scale. Pilots frequently lack auditability, policy controls, and human oversight required for high-risk workflows involving PHI. Additionally, errors emerge late in real-world environments, especially in claims and billing, where outputs fail payer rules or interoperability expectations. Fragmented technology ecosystems, including legacy systems, proprietary platforms, and uneven HL7/FHIR adoption, make integrations brittle and difficult for generic AI solutions.

Scaling succeeds only when AI aligns with real operational goals, is supported by strong data foundations, and is designed for complex workflows. A shift is now emerging toward focused, value-chain-driven MVPs targeting high-impact use cases with clear business value and executive sponsorship, moving the conversation from AI experimentation to measurable process transformation.

From a systems perspective, what does a modern AI-ready healthcare architecture look like, particularly in terms of data pipelines, interoperability, and cloud infrastructure?

An AI-ready architecture is one where policy enforcement, validation, and escalation are embedded directly into workflows, not managed externally as an afterthought. It requires unified cloud-native data platforms that ingest claims, EHR, imaging, device, and operational data into highly governed layers.

We must prioritize standards-first interoperability using FHIR, HL7, SMART on FHIR, and DICOM, supported by dedicated validation engines. There has to be a clear separation of concerns across ingestion, processing, analytics, AI services, and governance layers. Context engineering sits at the intersection of these layers; it is the discipline that connects raw ingested data to governed, semantically enriched inputs that AI services can act on with precision and accountability.

Built-in security is non-negotiable, encompassing RBAC, encryption, consent management, lineage, and audit trails. Architecture success is ultimately measured by whether systems fail safely, not by component sophistication. By encoding frameworks like HIPAA and GDPR directly into the execution layer, we build the architectural trust required for global deployment.

How are organizations handling real-time data processing and integration across electronic health records (EHR), medical devices, and payer platforms to enable AI-driven decision-making?

Real-time processing matters only when insights surface inside clinician and operator workflows, not downstream in secondary dashboards. The core challenge is managing variability and exceptions, rather than purely focusing on data throughput. Organizations are addressing this through middleware that normalizes and enriches data before AI consumption, rather than pushing raw feeds downstream.

This is where context engineering plays a critical role. The system must structure incoming data from EHRs, medical devices, and payer platforms into a coherent contextual foundation, ensuring that AI inputs are semantically aligned with the specific clinical or operational decision. This enables tight integration with EHR workflows, allowing insights to appear at the point of care.

More importantly, this makes integration meaningful rather than purely technical, ensuring outputs reflect not just data, but also clinical guidelines, payer rules, and workflow constraints relevant to each patient and encounter. The focus is on intercepting the workflow at the right moment and providing actionable support without disrupting existing operational rhythms.

What are the biggest technical challenges in deploying AI models in clinical environments, especially around model validation, monitoring, and drift management?

The hardest challenge is not accuracy decay alone, but undetected error propagation across interconnected workflows. Context engineering serves as the first line of defense by structuring the knowledge environment in which models operate. The system must validate inputs for clinical and operational coherence before AI reasoning begins, reducing errors at the source.

Validation must go beyond automated metrics and include human reinterpretation, ensuring outputs reflect real clinical relevance rather than just statistical performance. This approach grounds model evaluation in actual clinical and administrative outcomes instead of abstract benchmarks.

Model drift must also be actively managed as patient populations, clinical guidelines, and behaviors evolve. This requires continuous monitoring tied to real-world feedback loops, with checkpoints embedded to detect drift in terms of clinical relevance, payer rule alignment, and workflow consistency.

Ultimately, success depends on balancing adaptability with strict regulatory expectations. AI deployment in clinical and SaMD environments requires a strong trust layer, one that enforces guardrails, triggers human review, and ensures controlled retraining before any impact reaches patient care.

How do you approach building explainability and auditability into AI systems that are used in regulated healthcare settings?

Explainability exists so humans can contest, override, and learn from AI outputs, not just understand them. Context engineering makes this contestability structural rather than superficial by encoding clinical guidelines, policy rules, and workflow logic directly into the AI environment. As a result, every output can be traced back to the contextual inputs that shaped it, rather than being reverse-engineered after the fact.

Auditability, similarly, must create institutional memory rather than serve as a compliance afterthought. This is achieved by using policy-as-code to embed regulatory and organizational rules directly into execution flows, treating frameworks like HIPAA and CMS guidelines as governing inputs from the start, not external checks applied later.

In addition, immutable logs and audit trails are essential to support regulatory oversight and clinical trust. The system should capture not just what decision was made, but the full context, including data, constraints, and workflow state that informed it.

Across all of this, human-in-the-loop oversight remains an essential requirement for high-risk or irreversible clinical decisions. Building this transparent trust layer ensures that when a regulator or clinician asks why an AI system made a specific recommendation, the answer is immediately accessible and defensible.

With the rise of agent-based and autonomous systems, what safeguards are needed to ensure reliability and prevent unintended outcomes in clinical workflows?

In healthcare, agentic systems are valuable only when automation boundaries are explicit and reversible. Context engineering helps define these boundaries by structuring operational scope, clinical constraints, and escalation logic into the environment, ensuring autonomy operates within a governed framework rather than relying on models to self-regulate.

This requires autonomy with accountability, including clear role definitions, escalation paths, and deterministic checkpoints embedded into agent behavior. The system must ensure agents operate within clinically and operationally approved boundaries that are architecturally enforced, not just policy-defined.

It also requires continuous monitoring of agent behavior, decision quality, and unintended interactions. We implement policy-driven constraints that prevent agents from acting outside their approved clinical or operational scope. For clinical workflows, human-in-the-loop is a fundamental design principle, not a fallback mechanism.

As we move toward advanced agent-to-agent orchestration, these agents must operate strictly within governed, codified frameworks to ensure reliability.

Looking ahead, what does intelligence-driven care actually look like for patients over the next decade?

The next decade will see AI reduce friction before it transforms care, starting with administration, coordination, and decision support. Intelligence-driven care succeeds when clinicians trust the defaults but firmly retain their authority. AI becomes an orchestrator across the care continuum, anticipating needs rather than simply reacting to events.

Context engineering enables this shift by structuring longitudinal clinical, claims, and operational data so AI can reason across the full continuum with the depth required for reliable action, not just predictive insight. This supports more personalized, continuous, and context-aware care pathways across settings.

The phrase “context-aware” is the operative one, and it is not accidental. Delivering truly personalized care pathways at scale requires that AI systems inherit deep contextual knowledge of each patient’s clinical history, payer environment, and care setting. Clinicians will be supported by copilots and decision intelligence, significantly reducing their cognitive and administrative burden.

Over time, healthcare systems will evolve into learning systems, improving continuously as data, models, and real-world feedback compound. By engineering a robust trust layer today, we lay the operational foundation for this frictionless future.

Thank you for the detailed answers, readers who wish to learn more should visit CitiusTech.