Worked on enhancing the pymc-labs/pymc-marketing repository by extending the MMM class to support arbitrary HSGP instances for time-varying intercepts and media, enabling more flexible and accurate marketing-mix modeling. Focused on robust configuration and persistence, the developer implemented serialization and deserialization logic using Python, allowing HSGP configurations to be reliably saved and restored across sessions. Expanded unit test coverage ensured that custom HSGP serialization and deserialization worked as intended and prevented regressions. The work leveraged skills in Python programming, data modeling, and statistical analysis, resulting in a more adaptable modeling framework for campaign analysis without introducing new bugs.