Miloš Švaňa discusses the difficulties of setting up a PyTorch project that functions across various operating systems and hardware accelerators. He explores solutions using PEP 508 for dependency management and ultimately decides to switch from PyTorch to ONNX Runtime for easier installation and better compatibility with PyPI.

Bazel, despite its promise of hermeticity and reproducibility, presents significant challenges, particularly due to its read-only sandboxing and lack of robust Windows support. The article discusses three main "sins" of Bazel, including its dependency management issues and the complications arising from its attempts to cater to a broader user base, ultimately questioning the effectiveness of its approach compared to more curated systems.