As modern astrophysics confronts vast archives of observational data, the NASA AI ML STIG equips researchers with practical computational literacy. Established under the NASA Cosmic Origins Program as the Artificial Intelligence and Machine Learning Science and Technology Interest Group, the community initiative replaces unguided experimentation with structured training. Christopher Stubbs of Harvard University led the first dedicated practical session on September 21, 2026, initiating hands-on demonstrations of agentic coding workflows for astronomical research [1, 2].
Practical Tools Enter the STIG Curriculum
NASA Science scheduled the September 21 presentation under the formal title Hands-On Session I: Agentic Coding and Research Tools. Harvard University physicist Christopher Stubbs served as the featured instructor for the virtual gathering, presenting operational strategies for integrating generative code assistants into daily research pipelines. The event provided direct demonstrations of software tools rather than theoretical overviews. Agentic coding techniques allow astronomical researchers to automate repetitive script generation, parse complex telescope headers, and manage multi-step data pipelines. Both sessions met online [1, 2].
The virtual session was free [1]. Participants accessed the live demonstration without registration fees, joining colleagues worldwide through NASA Science digital links [2].
While PerEXP Teamworks previously examined theoretical models in its analysis of what NASA AI ML STIG teaches on agents, the September 21 event departs from conceptual architecture to demonstrate operational code execution [1, 4]. A subsequent follow-up titled Hands-On Session II: Agentic Coding and Research Tools features Serat Saad from The Ohio State University on September 28, 2026. By separating foundational mechanics from hands-on terminal applications across consecutive Mondays, the NASA AI ML STIG lecture series establishes a progressive learning trajectory designed specifically for practicing astronomers [1, 3].
Why NASA Prioritizes Domain AI Literacy
The NASA Cosmic Origins Program established the NASA AI ML STIG in late 2025 in response to accelerating computational demands across astrophysics. Observational facilities now generate petabytes of raw pixels, spectral cubes, and time-domain alerts that exceed manual inspection capacities. How can astronomers process petabyte-scale observations without specialized machine learning fluency? Without domain-specific machine learning skills, research teams risk falling behind the capabilities enabled by contemporary hardware and foundation models. The interest group provides structured, domain-specific instruction through modular training units (stackable, bite-sized curriculum units) specifically configured for astronomical research contexts [5].
Upskilling the astronomical workforce strengthens NASA’s competitive advantage in AI-enabled space science across international research consortia. Rather than relying on generic computer science coursework that overlooks observational noise and instrument systematics, the community initiative designs tutorials around concrete astronomical challenges. The program builds an interdisciplinary workforce capable of steering next-generation space missions while establishing institutional guidelines for responsible artificial intelligence adoption. Domain-specific training modules ensure that astronomical domain expertise guides model development [5].
Curriculum development across the interest group reflects a broader shift toward interactive computational training [4]. The ongoing series addresses specific astronomical applications through short tutorials, community town halls, and foundational modules designed for broad institutional reach [5]. Weekly talks recur every Monday [7]. NASA positions this community-driven structure as a scalable model for other agency science divisions facing parallel technical hurdles [5].
Open Textbooks Anchor Astronomy Machine Learning
A central pillar of the educational infrastructure created by the interest group is Deep Learning for Astrophysics: An Open Textbook from the NASA Cosmic Origins AI/ML Science and Technology Interest Group, compiled by a broad international collaboration. Co-chairs Yuan-Sen Ting from The Ohio State University and Digvijay Wadekar from the University of Texas at Austin led the project with Phill Cargile, Caroline Cuesta-Lazaro, Austin Curtis-Trudel, Gregory Green, Robert McClelland, Daniel Muthukrishna, Tri Nguyen, Tomasz Rozanski, Anna Scaife, Jesse Thaler, Licia Verde, Francisco Villaescusa-Navarro, Sihan Yao, Alex Gagliano, and Swara Ravindranath. Released in 2026 as an open preprint through arXiv at Cornell University, the textbook serves as an evolving living document curated directly from community lectures and technical workshops. The text awaits formal review [8].
Because the textbook manuscript remains an unreviewed preprint that has not undergone formal academic peer review, the authors emphasize practical reproducibility over definitive theoretical claims. The educational volume provides executable Jupyter notebooks, documented code examples, and structured exercises hosted publicly on GitHub and ai4astro.org [5, 8]. Students and researchers can clone these repositories to run machine learning algorithms on real telescope datasets without configuring custom deep learning environments from scratch [8].
NASA AI ML STIG educational materials integrate standard frameworks including PyTorch and JAX to illustrate astronomical data processing pipelines. By grounding abstract machine learning concepts in operational astrophysical problems, the curriculum bridges the persistent gap between computer science literature and day-to-day telescope data analysis [5, 8]. Community members contribute new tutorials following periodic STIG town halls, ensuring that emerging techniques remain accessible to early-career investigators [5].
Open Science Directives Shape Agency Infrastructure
Educational initiatives across the NASA AI ML STIG operate within strict agency data governance standards established during the 2023 Year of Open Science. Steven M. Crawford from NASA Headquarters presented these institutional frameworks in an earlier STIG seminar titled Open Science and AI at NASA, speaking on behalf of the Open Science Implementation team within the Science Mission Directorate’s Office of the Chief Science Data Officer. Crawford detailed agency policies designed to expand the availability, utility, and security of scientific research data across all astrophysics missions [6].
Agency directives codified under Scientific Policy Directive 41a mandate open-access publications, fully accessible research datasets, shared software source code, and transparent scientific meetings across NASA funded research programs. Crawford outlined core services supporting these mandates, including modernized computing capabilities provided through the agency’s Science Cloud and High-End Computing portfolios. Cross-division platforms like Science Explorer and the Science Discovery Engine facilitate interdisciplinary discovery across extensive mission archives [6].
To assist researchers in navigating these statutory requirements (SPD-41a), the Transform to Open Science initiative deployed the Open Science 101 curriculum and official NASA Open Science Certificate programs. These training modules reinforce reproducibility, ethical attribution, and data stewardship across every computational project. Open science requires reproducible workflows. Institutional alignment ensures that artificial intelligence pipelines deployed within cosmic origins research adhere strictly to federal transparency benchmarks [5, 6].
Agentic Workflows Expand Scientific Software Work
The conceptual foundation underlying automated research tooling appears in The Future of Artificial Intelligence and the Mathematical and Physical Sciences, a collaborative community report authored by Andrew Ferguson, Michael LaFleur, Lars Ruthotto, Jesse Thaler, Yuan-Sen Ting, Pratyush Tiwary, and Soledad Villar. Released as an unreviewed preprint on arXiv in September 2025, the synthesis highlights the necessity of domain-informed reasoning algorithms [9]. Early exploratory seminars led by Francisco Villaescusa-Navarro of the Flatiron Institute introduced multi-agent collaboration, hypothesis generation, and structured code generation to astronomical researchers [4].
Can autonomous agent routines reliably draft telescope scripts without hallucinating observational parameters? Francisco Villaescusa-Navarro demonstrated operational architectures based on ReAct (Reason + Act) patterns and LangGraph execution workflows to coordinate complex analysis pipelines. Community organizers emphasize that agentic code generators serve as interactive research assistants rather than autonomous substitutes for scientific reasoning. Resources hosted on ai4astro.org enable astronomers to test agent prompts on calibrated observational data [4, 5].
Future NASA AI ML STIG modules announced by co-chair Yuan-Sen Ting will explore retrieval-augmented generation (RAG) memory systems, tool execution through the Model Context Protocol, and mechanistic interpretability of trained neural networks [4, 5]. These advanced methods connect directly with long-range strategic goals outlined in NASA Cosmic Origins planning for future space observatories, where automated data handling will prove decisive. As space observatories collect increasingly dense data volumes, agentic frameworks may streamline routine pipeline operations across academic facilities [5, 9].
Community Access Drives NASA AI ML STIG
The governance of the NASA AI ML STIG relies on active community steering led by co-chairs Yuan-Sen Ting of The Ohio State University and Digvijay (Jay) Wadekar of the University of Texas at Austin. The leadership council brings together early-career researchers, including Alex Gagliano from MIT, Artem Poliszczuk from Stanford, Tri Nguyen from Northwestern University, Ce Sui from The Ohio State University, and Julie Rolla from the NASA Jet Propulsion Laboratory (JPL). Senior academic consultants Kelle Cruz of the City University of New York, Bhuvnesh Jain of the University of Pennsylvania, and Moritz Munchmeyer of the University of Wisconsin Madison provide strategic oversight [5].
The interest group actively invites graduate students, postdoctoral fellows, and early-career researchers to nominate themselves or colleagues to join the leadership council [5]. All weekly lectures remain open to the public without cost, streaming every Monday at 4:00 p.m. Eastern Time (1:00 p.m. Pacific). Public lecture recordings and tutorial notebooks are archived openly to ensure equitable international access across the astronomical community [5, 7]. Interested scientists can join the community distribution list by emailing AI-ML-STIG-join@lists.nasa.gov with the subject join [5].
NASA AI ML STIG weekly seminars demonstrate how targeted community education bridges complex machine learning developments and observational astrophysics [5, 7]. Code verification remains an open problem. Rather than presenting finished algorithms as definitive solutions, organizers continue short tutorials and town hall discussions to refine emerging scientific tools against rigorous astronomical benchmarks [5, 7].
- WEBSITE Tyler, P., & Cosmic Origins Team. (2026, September 21). AI/ML STIG lecture series, 21 Sept 2026. NASA Science. [Article Link]
- WEBSITE Tyler, P., & Cosmic Origins Team. (2026, September 16). AI/ML STIG seminar, 21 Sept 2026. NASA Science. [Article Link]
- WEBSITE Tyler, P., & Cosmic Origins Team. (2026, September 21). AI/ML STIG lecture series, 28 Sept 2026. NASA Science. [Article Link]
- WEBSITE Tyler, P., & Cosmic Origins Team. (2026, September 14). AI/ML STIG lecture series, 14 Sept 2026. NASA Science. [Article Link]
- WEBSITE Tyler, P., & Cosmic Origins Team. (2025, November 18). Artificial Intelligence & Machine Learning Science & Technology Interest Group (AI/ML STIG). NASA Science. [Article Link]
- WEBSITE Crawford, S. M., Tyler, P., & Cosmic Origins Team. (2026, June 29). AI/ML STIG lecture series, 29 June 2026: Open Science and AI at NASA. NASA Science. [Article Link]
- WEBSITE Tyler, P., & Physics of the Cosmos Team. (2026, August 27). AI/ML STIG weekly talks starting 14 Sept 2026. NASA Science. [Article Link]
- PREPRINT Ting, Y., Wadekar, D., Cargile, P., Cuesta-Lazaro, C., Curtis-Trudel, A., Green, G., McClelland, R., Muthukrishna, D., Nguyen, T., Qu, H., Rozanski, T., Scaife, A., Thaler, J., Verde, L., Villaescusa-Navarro, F., Wu, J. F., Xu, D., Yao, S., Gagliano, A., … Ravindranath, S. (2026). Deep Learning for Astrophysics: An Open Textbook from the NASA Cosmic Origins AI/ML Science and Technology Interest Group. arXiv (Cornell University). [Article Link]
- PREPRINT Ferguson, A., LaFleur, M., Ruthotto, L., Thaler, J., Ting, Y., Tiwary, P., Villar, S., Alves, E. P., Avigad, J., Billinge, S., Bilodeau, C., Brown, K., Candes, E., Chattopadhyay, A., Cheng, B., Clausen, J., Coley, C., Connolly, A., Daum, F., … Frutos, L. M. (2025). The Future of Artificial Intelligence and the Mathematical and Physical Sciences (AI+MPS). arXiv:2509.02661. [Article Link]
APA 7: TWs Editor. (2026, September 21). NASA AI ML STIG Opens Agentic Coding Series for Astronomy. PerEXP Teamworks.
