Reproduce “Reliability by Architecture: Agentic Workflows for Validated Multi-Physics Simulation from Natural Language” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Søren Taverniers, Eric Darve, Hamdi A. Tchelepi
The authors present a workflow for preparing and checking geologic CO₂-storage simulations, assigning request interpretation and uncovered components to an LLM and construction and verification to deterministic routines. Grounding alone raised schema-valid decks from 5/20 to 20/20 but yielded no accepted runs; adding synthesis and normalization produced 20/20 accepted runs in the tested flow tasks, not a general success rate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.