Worked on the emcie-co/parlant repository to deliver a Gemini Embedding Service model upgrade, migrating from text-embedding-004 to gemini-embedding-001. This involved refactoring the service in Python to update model names and configurations, reducing the maximum token count from 8000 to 2048 and increasing embedding dimensions from 768 to 3072. The approach emphasized configuration management and API integration, enabling richer embeddings while controlling latency and compute costs. By establishing a maintainable, configuration-driven upgrade path, the work improved the service’s adaptability for future model changes and reduced deployment risk, focusing on feature delivery and long-term maintainability over bug fixes.