Worked on the castai/helm-charts repository to deliver a targeted feature enhancing cache control in hosted-model deployments. Developed support for the kv-cache-dtype parameter within the vLLM sub-chart, allowing users to fine-tune key-value cache data types for improved performance and memory management. The implementation involved updating YAML deployment manifests, Helm chart versions, and documentation to clearly surface the new configuration option and its operational impact. Ensured backward compatibility and adherence to repository standards throughout the process. Leveraged expertise in Helm, Kubernetes, and YAML to provide a maintainable, well-documented solution that aligns with established versioning and deployment practices.