A community-driven forum advancing principled approaches to building LLM systems that can be systematically checked, audited, and trusted — rather than assumed correct.
Correctness in LLM training, serving, and post-training has emerged as a central bottleneck to further advances in model capability, reliability, and deployment at scale.
These challenges stem from the inherent stochasticity of learning algorithms and the pervasive nondeterminism in modern LLM systems — amplified by rapidly evolving hardware, distributed execution, and complex software toolchains. As a result, today’s LLM systems often rely on ad hoc testing, heuristics, and trust in opaque components.
RigoLLM brings together researchers and practitioners from systems, machine learning, verification, and programming languages to examine foundations, tools, and methodologies for rigor in LLM systems. It is a forum to articulate open problems, share early results, and shape a research agenda toward LLM systems that can be systematically checked, audited, and trusted.
Prioritizing formalism, systematic methods, and principled design over ad hoc heuristics.
Formally defined problems with explicit correctness specifications — making LLM systems checkable and analyzable as engineered systems.
We call for submissions on topics including, but not limited to:
Authors may submit in one of the following categories. All accepted submissions are encouraged to present a poster to facilitate interaction and discussion.
Work in the broad area of rigorous LLM systems. Submissions may present completed techniques or novel and promising ideas that require further development.
Researchers and practitioners are invited to describe early-stage work that is not yet ready for full publication.
Submissions that revisit previously published results from a new perspective, or describe extensions of prior work by the authors.
Evaluation. Submissions are evaluated on originality, technical merit, topical relevance, value to the community, and potential to stimulate insightful discussion. Submissions are kept confidential. Authors may choose whether to include their papers in the proceedings, to encourage open discussion and the presentation of early-stage ideas.
All dates are tentative and given in AoE. The workshop takes place during ACM ATC 2026 (Nov 15–18, 2026).
Reach out to the primary contact for any questions about RigoLLM 2026. For venue, registration, and travel details, see the co-located conference website linked below.
Cheng Tan · Northeastern University
For venue, registration, and travel, see the ACM ATC 2026 website.