Leonid Stolbov 1, Kirill Zhudenkov 1,2, Victor Sokolov 1,3
1 M&S Decisions (Dubai, Arab Emirates), 2 Research Center of Model-Informed Drug Development, Sechenov First Moscow State Medical University (Moscow, Russia), 3 Marchuk Institute of Numerical Mathematics of the Russian Academy of Sciences (Moscow, Russia)
Introduction
Pharmacometrics analyses typically require modelers to apply multiple methodologically distinct functions across diverse tools and environments, prompting substantial reformatting efforts and increasing both time costs and error risks – motivating the development of more integrative, user-friendly platforms [1,2]. The rapid evolution and growing accessibility of AI tools has opened a fundamentally new layer of opportunities for building such environments, spanning both scripted and graphical user interface (GUI)-based approaches [3,4]. In this work we present Simurg – an ecosystem for applied pharmacometrics that integrates three complementary interface paradigms: script-based workflows in R and Python, an application-driven infrastructure with a GUI, and local human-in-the-loop AI assistance.
Methods
Simurg operates as a unified, governed ecosystem supporting both Linux/Windows desktop deployment and a web-based TypeScript application with a plugin architecture. The backend is built on a Node.js server capable of interfacing with SQL databases, executing C++ algorithmic code, and orchestrating Docker containers to run R and Python scripts as well as RStudio Server. Plugins are distributed as Webpack packages that can be installed or removed on demand; each exposes an intuitive GUI with component-level documentation powered by Storybook. AI support is embedded through a locally deployable assistant, Sirin AI, which is composed of multiple large language models operating as a coordinated agent pipeline.
Results
The Simurg ecosystem achieves a unique combination of capabilities through the integration of scripting, GUI and AI. First, it enables seamless interaction between the GUI and scripting languages: model development conducted within the ecosystem can be readily exported to R scripts, allowing full reproduction of the workflow using both the GUI and the SimuRg package.
Second, the architecture supports plugin and GUI development without requiring programming expertise, by exposing core GUI components as reusable building blocks and providing shared algorithmic and scripting modules. This design offers a unique opportunity for the broad community of pharmacometricians, including specialists without software development background, to integrate novel mathematical methods, models, and functionalities by building and deploying new applications and gradually expand the ecosystem. Three plugins are currently available that cover all key steps of a NLME workflow: (1) EVA enables automatic data extraction from time-series and scatter plots, and provides tools for evaluation, correction, and downstream analysis of the extracted results; (2) NLME plugin serves as a unified API for conventional population modeling tools (nlmixr2, Monolix) while also incorporating its own C++ core for ODE solving and parameter estimation; (3) RECORD streamlines the preparation of reproducible modeling reports in key formats including R Markdown, DOCX, and PPTX.
Finally, Sirin AI enables rapid onboarding, code linting and modification, as well as contextual support through natural-language interaction. It is aligned with regulatory principles for AI in drug development – including explicit context of use, risk-based oversight, data governance, lifecycle management, and transparency [5,6] – and is further augmented by an internal knowledge base containing documentation, architecture descriptions, and domain-specific references.
Conclusions
Simurg is an ecosystem for pharmacometricians that unifies three paradigms – script-based workflows, an application-driven GUI infrastructure, and local human-in-the-loop AI assistance – into a coherent environment for executing end-to-end modeling workflows. Future directions include development of a more specialized AI agent capable of directly invoking ecosystem functions, expansion of no-code tooling for custom plugin creation, and integration of additional intelligent modules to further support analysts across the modeling lifecycle.
References:
[1] P. C. Marshall et al. (EFPIA MID3 Workgroup), “Good Practices in Model-Informed Drug Discovery and Development: Practice, Application, and Documentation,” CPT: Pharmacometrics & Systems Pharmacology, 2016, doi: 10.1002/psp4.12049.
[2] J. J. Wilkins et al., “Thoughtflow: Standards and Tools for Provenance Capture and Workflow Definition to Support Model-Informed Drug Discovery and Development,” CPT: Pharmacometrics & Systems Pharmacology, vol. 6, no. 5, pp. 285–292, May 2017, doi: 10.1002/psp4.12171.
[3] Peng, S., Kalliamvakou, E., Cihon, P., and Demirer, M. “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot.” arXiv (2023). doi: 2302.06590.
[4] Ajimati, M. O., et al. “Adoption of low-code and no-code development: A systematic literature review and future research agenda.” The Journal of Strategic Information Systems (2025). (ScienceDirect record; SLR on LCNC adoption / citizen development / accessibility beyond professional developers). doi: 10.1016/j.jss.2024.112300
[5] D. B. Olawade, S. C. Fidelis, S. Marinze, E. Egbon, A. Osunmakinde, and A. Osborne, “Artificial intelligence in clinical trials: A comprehensive review of opportunities, challenges, and future directions,” Int. J. Med. Inform., vol. 206, p. 106141, 2026, doi: 10.1016/j.ijmedinf.2025.106141.
[6] European Medicines Agency (EMA) and U.S. Food and Drug Administration (FDA), “Guiding principles of good AI practice in drug development,” Jan. 2026.
Reference: PAGE 34 (2026) Abstr 12079 [www.page-meeting.org/?abstract=12079]
Poster: Methodology – AI/Machine Learning