CoLLM: AI engineering toolbox for end-to-end deep learning in collider analyses
W. Esmail, A. Hammad, M. Nojiri
TL;DR
CoLLM presents a physics-aware, end-to-end framework that uses LLMs to generate executable collider-analysis code from plain-language specifications and to automate downstream deep learning, all within a reproducible, plug-and-play workflow. It couples a physics-constrained code-generation engine with automated DL pipelines (MLP, GNN, Transformer) and a LangChain-based orchestrator that enforces deterministic decoding and iterative error correction. The approach is validated on five benchmark processes, demonstrating executable analyses, validated observables, and DL classifiers, while highlighting comparative advantages over generic LLM code generation. The work aims to substantially reduce coding and technical overhead in collider analyses while preserving physical reliability, with public availability and a clear roadmap for future extensions such as ROOT-format support and adaptive graph constructions.
Abstract
Recent improvements in large language models have opened new opportunities for accelerating and automating scientific workflows. In parallel, modern collider analyses are becoming increasingly complex and demand substantial programming and deep learning expertise. \coll alleviates this workload by using pretrained large language models to generate physically consistent analysis code for event selection. Additionally, it automates subsequent deep learning analyses. To further reduce reliance on programming or deep learning experience, \coll provides a graphical user interface that allows users to perform end-to-end analyses through an interactive interface. The main motivation behind \coll is to lower the coding burden and simplify the technical complexity of collider analyses, which increasingly depend on sophisticated event selections and advanced deep learning methods.