.png){kind=logo}
Building Trustworthy AI for Public Health: Advancing Agentic AI, Data Modernization, and Responsible Innovation
Artificial intelligence is transforming how organizations analyze data, generate insights, and support decision-making, but in high-stakes environments such as public health, success depends on far more than model performance alone. As governments and healthcare organizations increasingly adopt large language models, agentic AI, and advanced analytics, the focus is shifting toward building systems that are transparent, governed, measurable, and capable of delivering reliable results at scale.
In this interview, Anindita Nath discusses how advances in generative AI, biomedical informatics, and data modernization are reshaping public health and scientific research. With more than a decade of experience developing AI and machine learning solutions across industry and research, they have led initiatives spanning large language models, natural language processing, metadata engineering, and AI-ready data infrastructure. Their work includes co-leading one of the first agentic AI deep-research evaluation initiatives in the federal public health space, helping establish early frameworks for the responsible evaluation and adoption of autonomous AI systems.
Drawing on experience building AI systems for disease surveillance, biomedical research, and large-scale public health data platforms, Nath shares insights into the challenges of governing AI in mission-critical environments, modernizing legacy data systems, and ensuring that intelligent technologies augment human expertise rather than replace it. They also explore the future of trustworthy AI, explaining why robust evaluation, transparency, and interdisciplinary collaboration will be essential as AI becomes increasingly embedded in healthcare and public health operations.
Please tell us about your professional journey and what inspired you to pursue a career in data science and artificial intelligence?
My career began in software engineering and enterprise IT digitization, then expanded into computer science research in natural language processing, speech and audio signal processing, statistical machine learning, deep neural networks, and biomedical informatics. I was particularly drawn to AI’s potential to reduce cognitive load in operational environments. My early work in conversational AI and assistive health technologies shaped my view that intelligent systems should go beyond text generation to support information retrieval, knowledge synthesis, and decision support. This perspective continues to guide my work in public health, where AI and modern data engineering can strengthen surveillance, accelerate scientific discovery, and improve evidence-based decision-making.
Your work spans AI, public health, biomedical informatics, and data modernization. How would you describe your current role and areas of focus?
My current role centers on designing governed, production-oriented AI systems for public health intelligence. My work spans generative AI, large language models, agentic AI enabled system evaluation, disease surveillance, metadata engineering, and AI-ready data infrastructure. I convert fragmented public health and biomedical datasets into structured, provenance-aware assets that support analytics and decision-making. A core priority is ensuring that these systems are measurable, reproducible, privacy-preserving, and operationally useful for analysts, researchers, and public health leaders.
Over the past decade, what have been some of the most defining moments or achievements in your career?
Several achievements have shaped my professional trajectory. I co-led the first agentic AI deep-research evaluation initiative in federal public health, integrating advanced AI capabilities with governance, safety assessment, and operational validation to inform the agency’s agentic AI adoption strategy. I also led GENEVIC, an Azure OpenAI-enabled biological copilot for genomic data exploration, improving access to complex genomic information while preserving scientific rigor. Additional accomplishments include developing AI-enabled metadata and surveillance tools that reduced manual review burden by almost 50%, publishing first-author research in Bioinformatics, AMIA, ACM, and Speech Prosody venues, mentoring learners across computing and AI, receiving the Generation Google Scholarship in 2020, and being featured among Influential Women in 2026 and several such recognitions that reflect sustained technical performance, leadership, and impact.
For readers who may be new to your work, what excites you most about the intersection of AI and public health today?
I am most interested in AI’s capacity to move public health from retrospective, static reporting toward interactive, evidence-driven intelligence. Public health decision-making, for example, in disease surveillance monitoring requires integrating heterogeneous signals from surveillance systems, clinical encounters, laboratory testing, immunization records, demographic datasets, scientific literature, and policy guidance. AI can accelerate signal integration and knowledge synthesis when supported by high-quality data, transparent methods, robust validation, and expert oversight. Its value is not to replace public health judgment, but to augment experts’ ability to detect patterns earlier, evaluate uncertainty, and act on stronger evidence.
You co-led the first agentic AI evaluation initiative in the federal public health space. What motivated this effort, and what were some of the key learnings from the project?
The motivation was clear: before autonomous or semi-autonomous AI systems can support high-stakes public health work, they must be carefully evaluated. We needed to understand how agentic AI performs across complex questions involving epidemiology, informatics, policy guidance, and legal review. A key learning was that accuracy alone is not enough. Evaluation must include relevance, completeness, source reliability, safety, reproducibility, compliance screening, human subject-matter expert review, documentation of model behavior, source verification, and explicit identification of limitations. Governance must be designed from the beginning, not attached later.
As organizations increasingly deploy autonomous AI systems, how should leaders approach the challenge of evaluating and governing these technologies in high-stakes environments?
Leaders should treat evaluation as an operational requirement, not a one-time technical test. Autonomous AI systems need clear task boundaries, trusted data sources, audit trails, privacy safeguards, human review, and continuous monitoring. A public health AI system should know whether it is retrieving evidence, summarizing a report, analyzing data, drafting communications, supporting policy planning, or identifying limitations. In high-stakes environments, leaders must ask not only whether a model can answer, but whether the organization can trust, verify, reproduce, and govern that answer.
You have worked extensively on AI-driven data modernization. What are the biggest challenges public health organizations face when transforming legacy data systems into AI-ready infrastructure?
The biggest challenge is fragmentation. Public health knowledge is often trapped in isolated spreadsheets, inconsistent field definitions, separate dashboards, different reporting timelines, and legacy systems that were not designed for AI. Before AI can be useful, organizations need strong foundations: standardized data models, metadata, provenance, quality checks, access controls, documentation, interoperability, and reliable pipelines. Data modernization is also a workflow and trust challenge. AI-ready infrastructure requires clear accountability so that every output can be traced back to the evidence that produced it.
Your work has involved leveraging Azure OpenAI, Databricks, and Palantir Foundry for large-scale metadata analysis and knowledge discovery. Can you share a real-world example where these technologies delivered measurable impact?
One example is my work on LLM-enabled metadata exploration and visualization. Through systems such as MetaMation and related enterprise metadata tools, users could ask natural-language questions over complex health surveillance and program-inventory metadata, retrieve relevant information, and generate visual summaries. This reduced manual analyst effort by more than 50% and made siloed metadata more accessible to program managers. GENEVIC was another example: a grounded biological copilot that used curated data, APIs, secure cloud deployment, prompt optimization, and validation so genomic exploration could become more intuitive without sacrificing scientific integrity.
Looking ahead, how do you see large language models and agentic AI reshaping healthcare, public health surveillance, and government operations over the next five years?
Over the next five years, large language models and agentic AI will likely mature into governed orchestration layers across clinical care, public health surveillance, and government operations. In healthcare delivery, domain-specific agents could support ambient clinical documentation, EHR-integrated decision support, diagnostic triage, longitudinal patient summarization, prior authorization workflows, medication reconciliation, care-gap detection, remote monitoring, and precision-medicine recommendations.
More transformative strategies may include multimodal models that integrate clinical notes, imaging, laboratory results, genomics, wearables, and social-risk indicators to enable earlier risk stratification and more personalized intervention planning. In public health, similar systems can strengthen evidence retrieval, epidemiologic signal detection, literature synthesis, metadata discovery, automated reporting, and cross-jurisdictional situational awareness.
The most effective implementations will be embedded within secure informatics infrastructure rather than standalone chat interfaces. Their reliability will depend on grounding, provenance tracking, uncertainty estimation, continuous model monitoring, access control, interoperability, and human-in-the-loop review. These capabilities can expand expert capacity, reduce administrative burden, and improve operational responsiveness; however, accountability for interpretation, clinical action, ethics, equity, and final decisions must remain with human experts.
Beyond your technical contributions, you are actively involved in mentorship and STEM advocacy. What advice would you give to aspiring AI professionals looking to build impactful and purpose-driven careers?
My advice is to develop technical depth alongside a clear sense of purpose. Aspiring AI professionals should build strong foundations in programming, machine/deep learning, AI evaluation strategies, and technical communication while remaining attentive to the communities, and decisions their systems are intended to support. My own path has been shaped by perseverance, first-generation challenges, and a commitment to making computing both rigorous and accessible. Through initiatives such as Microsoft TEALS, Girls Who Code, CodePath, AnitaB.org, IEEE, Data for Impact, Women in STEM Network, and CAHSI-related workshops, I aim to help emerging technologists build confidence, strengthen their skills, and see themselves as contributors to the systems that will shape the future.
If you would like to add or highlight any other interesting insights that we might have missed in our questionnaire, please feel free to do so.
The next phase of AI leadership will require interdisciplinary fluency. Effective AI professionals must pair algorithmic expertise with domain knowledge, ethics, governance, communication, accessibility, and implementation skills. In public health, impact depends on turning technical capability into trusted, practical systems.
