Worked on backend enhancements for the UKGovernmentBEIS/inspect_ai and UKGovernmentBEIS/inspect_evals repositories, focusing on deployment reliability and evaluation tooling. Delivered Docker Compose configuration improvements, including memory management and expanded service key support, using Python and Docker to enable more flexible, validated deployments. Improved error handling by surfacing partial output on sandbox execution timeouts, aiding debugging and feedback. Addressed data management by repointing evaluation datasets to restore functionality. Extended the schema_tool_graded_scorer to support multi-field rubric payloads, allowing richer grading structures and better error handling. Emphasized robust API design, unit testing, and maintainable workflows throughout the development process.