As technology advances, the process of modernizing healthcare IT is shifting from simply replacing legacy applications and upgrading infrastructure. Even simple changes can be tricky to pull off in many health care organizations that are still using tightly coupled systems, databases that must be shared, polling, and nightly batch processing. Event-driven architecture is another way of doing it, which focuses on business events and notifies systems to communicate asynchronously, scale separately, isolate failure, and process information as close to the time it is created. The architectural change is also gaining significance as healthcare organizations are looking to new capabilities via artificial intelligence (AI) without impacting critical clinical workflows or compromising sensitive information.
Rajmani Singh has been working to overcome the architectural constraints of established healthcare systems. Singh has 18+ years of experience in software development, engineering, and enterprise architecture, where he advanced through various senior roles, from Software Developer II (2018) to Software Engineer Advisor (2023) and Solution Architect Advisor (2024). He has been increasingly focused on how to support complex healthcare environments to move away from database-based and batch-oriented architectures towards event-driven architectures that offer more flexibility, resilience and operational control.
As the years went on, Singh developed event-driven architecture as a pattern for event-driven modernization, as opposed to an application-specific solution. His solutions have involved publishing domain events, decoupling producers from consumers, minimizing direct database connections, providing support for asynchronous processing, adding retries, dead-letter queues, replay, idempotency, correlation, and observability features. This allows for more autonomous evolution of systems, and failure in one part will not affect other parts as much as a total failure would. This has been leveraged to process medical records, accept secure documents, abstract clinical data, perform quality control, integrate with FHIR, and develop applications.
One such example of this type of work is the movement of a clinical abstraction process from a batch nightly model to an event-driven process. Singh was involved in Project Meridian where he helped design a system where a case event kicks off retrieval of relevant FHIR and document information before processing proceeds through independently managed phases. Architecture consists of 6 distinct capabilities: case-event publishing, workflow orchestration, FHIR document retrieval, clinical processing, event-based output consumption and end-to-end observability. Importantly, the modernisation was planned to co-exist with the existing workflow, not to be an immediate replacement. The legacy process could run concurrently, enabling a selected pilot facility to be introduced to the new architecture, for reconciliation and validation prior to further migration.
Meanwhile, Singh's research has also broadened the scope of event-driven concepts into integrating AI systems. In the clinical abstraction pipeline, the AI process is not constrained to a set schedule that depends on servers, but rather, it is triggered based on individual records. This enables AI processing to grow without the healthcare workflow it is connected to. He also contributed to the development of a serverless AI processing model with real-time API ingestion, cloud-based model inference and asynchronous messaging in other parts of the world. Prompts are stored as versioned configurations to enable them to change over time without modifying the application code. Furthermore, having experience with a variety of cloud AI platforms has enabled him to take a provider-independent approach, with models being treated as replaceable consumers, instead of being locked in as a hard dependency inside of core applications.
But scalability and speed of development aren't all that can be used to judge the level of modernization in healthcare. The protection of protected health information (PHI), auditing, and the continuity of operations are all equally critical. The architectural choices Singh makes in her design for the system then include protection measures like encrypted event payloads, authentication, redaction of telemetry, prompt versions, model versions, human-in-the-loop approval, risk tiers, and security reviews. Likewise the workflows are designed from the beginning to include resilience mechanisms like retries, dead-letter handling, replay, buffering, reconciliation and correlation tracking. This means that an individual failure, such as failure of an AI service, can be contained without impacting the primary clinical transaction.
This approach has been seen mostly in the size and extent of the architectural transformation. Event-driven patterns have been used in over seven different healthcare technology areas, such as medical-record exchange, document intake, clinical abstraction, FHIR integration, AI-powered quality control, and controlled application platforms. The Project Meridian modernization was organized as a separate project with an estimated 136 story points, and it was planned to be phased to minimize the risk of operation. Simultaneously, Singh has contributed to an AI-assisted architecture, security, and production readiness engineering model that allows for a sustained rate of 1-2 pull requests per day while retaining a human component.
The experience does, however, emphasize a crucial difference: a move to an event broker or to AI does not equal modernization. Events should be a stable business concept and not a structure in the database; producers and consumers should have definite ownership. Similarly, AI systems must be governed, version controlled, protected from unauthorized access, and humanized for reliable integration into healthcare operations. Singh believes that observability, replay, security and phased migration are architecture requirements that should be considered in the initial stages.
In the future, there will be a continued trend of convergence of EDA, FHIR interoperability, cloud-native orchestration and AI-assisted processing in healthcare IT. Important is not only that technology gets faster, but that systems can adapt without continually having to do high-risk, large-scale replacements. Healthcare organisations can build a stepping stone for phasing in modernisation by decoupling business events from the underlying database and enabling AI capabilities to be independently scalable and observable. Singh's overall message is that there is a need to strike a balance between technology flexibility and reliability, governance and accountability, especially in contexts where sensitive healthcare information and critical clinical processes are being supported by the technology.
