The quest of particle physics is to infer the fundamental parameters of the Standard Model… or discover what might lie beyond. Machine learning has already begun to transform the HEP analysis landscape with the neural networks utilized across almost all stages of the data reconstruction pipeline and LHC physics analysis. Differentiable programming utilizes differentiable building blocks to build up a program (or analysis) whose components can be optimized jointly in an end-to-end fashion. The work in this seminar presents a new application for differentiable programming in particle physics. As our differential building blocks we consider (1) a reconstruction task (identifying boosted Higgs bosons) and (2) an analysis selecting di-Higgs events from multi-jet backgrounds. As particle-identification (such as Higgs-tagging) benefits from recent developments in foundation models, this amounts to fine turning such a foundation model for an analysis specific task. We demonstrate a factor of two increase in background rejection with these end-to-end pipelines showcasing an example of a heavy resonance decay to two Higgs bosons and four b-quarks from the Open CMS dataset, taking a ParT backbone as the foundation model. More details in https://arxiv.org/abs/2401.13536