Migrating Pydantic 1 to 2 with a generated build matrix (akasa.com)

AKASA's engineering team has successfully navigated the complex upgrade from Pydantic v1 to v2 by creating a shim layer that allows for gradual migration while maintaining compatibility across their extensive codebase. This transition, initially expected to be straightforward, revealed significant dependency coordination issues as various services were locked into incompatible library versions. By isolating the differences between Pydantic versions with their innovative shim library, they could write Pydantic 2 syntax while still operating on Pydantic 1. This approach not only minimized risks of regression in production but also allowed engineers to incrementally update services without the heavy overhead of a coordinated all-or-nothing upgrade. The significance of this accomplishment extends beyond just a library update; it highlights the often-overlooked intricacies of infrastructure management within AI systems, particularly in real-world applications dealing with complex data like healthcare. The development process included leveraging AI for generating an extensive compatibility test suite, which streamlined the verification of the shim implementation by quickly converting the Pydantic 2 migration requirements into executable tests. This not only demonstrated the potential for AI to reduce labor costs but also underscored its role in tackling technical debt within growing codebases, making the upgrade feasible without disrupting ongoing product features.