.png){kind=logo}
Artificial intelligence has moved beyond the stage where its promise is measured mainly through demos, benchmarks and controlled experiments. The harder test is whether AI can work inside real organisations, where data is incomplete, questions are ambiguous, users need context and decisions can affect people, operations and public trust.
Varun Kumar Nomula’s work is centered on that problem. An AI technologist and applied machine-learning researcher, Nomula works across applied AI, healthcare and public-health analytics, enterprise AI systems and AI governance, with a particular focus on how machine-learning systems perform outside controlled settings.
His work reflects a shift across the field. AI is becoming less of a standalone technical capability and more of an operating layer for decision-making. Organisations are using it to interpret language, retrieve information, automate analysis and support complex workflows. As AI becomes more useful, it also becomes more exposed to real-world complexity.
One part of Nomula’s work examines how AI can make sense of complex human language at scale. In public health, online conversations about vaccines can include concern, misinformation, confusion, personal experience, access barriers and genuine requests for clarification. Processing that volume of information is only part of the challenge. Understanding what it means is harder.
One of Nomula’s published research contributions, appearing in Human Vaccines & Immunotherapeutics, examined vaccine-related informational milestones, social-media discourse and administration trends for HPV and MMR vaccines in the United States. The research, supported by Merck, connected large-scale analysis of public discussion with real-world vaccination patterns and questions relevant to public-health communication and vaccine uptake.
The finding carries a broader lesson for AI. Large models and analytical systems can process public conversation at enormous scale, but language does not always translate neatly into intent or behaviour. A surge in vaccine discussion may reflect news coverage, concern, confusion, misinformation or access barriers. A negative post may not mean refusal. A question about safety may be a request for clarity rather than opposition.
For Nomula, this is where AI needs context: the technology can surface patterns, but public-health decisions still require interpretation, validation and human judgement.
Another research contribution involving Nomula, published in JMIR Formative Research, examined how language models performed on vaccine-related social-media content. The study showed the promise of using AI to process health communication at scale, while also exposing clear limits. Models could struggle with neutral statements, sarcasm, indirect concern and posts where confidence, access and practical barriers overlapped.
Those limitations point to a more demanding standard for applied AI: systems designed to support better decisions rather than simply produce faster outputs. Public language is full of ambiguity, and consequential settings require more than classification. They require context, validation and an understanding of when automated interpretation should support human decision-making rather than replace it.
The same issue appears in Nomula’s broader healthcare AI work. His published work includes research on AI-driven clinical decision support, including work presented through IEEE ICCCNT, as well as research involving AI-enhanced X-ray and MRI analysis, medical sensor data and healthcare analytics.
These projects use healthcare as a practical test case for a larger AI problem: what happens when model outputs influence decisions that require expertise, context and accountability?
In controlled settings, AI systems can appear highly capable. Real environments introduce more complexity: data can be incomplete, workflows can vary and users may ask questions in unexpected ways. A system trained or tested in one setting may behave differently in another. AI reliability therefore cannot be judged only by benchmark performance.
As systems become more general-purpose, the harder test is whether they can handle real-world context, recognise uncertainty, respect domain constraints and remain accountable when their outputs influence decisions.
Nomula’s applied enterprise AI work brings that issue into data access. He was a co-author of an article published on the AWS Machine Learning Blog documenting a collaboration between MSD, known as Merck & Co., Inc. in the United States and Canada, and the AWS Generative AI Innovation Center to implement a generative AI text-to-SQL solution for complex healthcare databases.
The work addressed practical challenges that emerge when generative AI moves beyond demonstrations into complex enterprise data environments. Effective text-to-SQL requires more than asking a language model to write code. Models need access to database structures, schema information and relevant context to generate executable, schema-specific queries.
Healthcare databases add another layer of difficulty. Questions may be vague, coded values may not be intuitive, and the system must preserve the user’s intent while navigating specialised data structures. The implementation therefore illustrates a broader point in enterprise AI: model capability alone is not enough. Architecture, context, validation and user oversight determine whether a system can be useful in practice.
A related example appeared in Nomula’s ISPOR 2026 work, published in Value in Health, on a real-world data large language model assistive SQL coding system. The project combined foundation models, retrieval-augmented generation, database metadata and verified examples, while emphasising user review, role-based access, benchmarking and further evaluation before broader use.
Together, these examples show that enterprise AI success depends on more than generating a technically plausible answer. Data quality, architecture, access controls, evaluation, monitoring and user oversight all matter. A system that gives users faster access to data must also help them determine whether the generated answer reflects their actual intent.
Nomula’s AI governance work brings the same concern to the organisational level. His focus is on how AI systems are reviewed, deployed, monitored and improved once they move beyond experimentation.
Rather than treating governance as paperwork, this work looks at the practical controls that determine whether AI can be used responsibly: ownership, human oversight, access control, evaluation, escalation and lifecycle monitoring.
That focus is becoming more important as organisations across the United States adopt AI in higher-impact settings. The early phase of AI adoption rewarded speed and experimentation. The next phase will reward systems that can be trusted under pressure: systems that make uncertainty visible, allow human review, protect sensitive information and continue to perform after deployment.
Across Nomula’s work, AI is treated as part of a larger decision system. His vaccine-related research focuses on interpreting complex public information and understanding the limits of automated interpretation. His healthcare AI work explores how intelligent systems can support decisions in sensitive environments. His enterprise AI work shows how generative AI can make complex data more accessible in real organisational settings. His governance work focuses on the structures needed to deploy such systems responsibly at scale.
That combination points to a broader shift in artificial intelligence.
More powerful models will continue to emerge, but technical capability alone will not determine their impact. What matters increasingly is whether AI systems improve decisions, respect context, handle uncertainty, support human judgement and remain accountable after deployment.
Varun Kumar Nomula’s work reflects that shift. It shows what real-world AI adoption increasingly requires: technical capability, domain understanding, practical governance and the discipline to make advanced systems useful where their outputs actually matter.
