Developed and delivered Bedrock Prompt Caching with model-type configuration for the block/goose repository, focusing on reducing redundant prompt processing and lowering latency for Anthropic Claude models. The solution introduced per-model-type caching controls and a configuration toggle, enabling staged rollout and experimentation while minimizing risk. Implemented entirely in Rust, the work leveraged backend development and API integration skills to provide flexible caching options that can be enabled or disabled as needed. This feature established a foundation for future performance metrics and cost savings by allowing more granular control over prompt caching, supporting ongoing optimization efforts for different model types.