• Turned a vague sponsor vision for ocean-monitoring device deployment and retrieval into an executable drone-based program by defining MVP scope, system architecture, and module ownership across navigation, perception, hardware, and telemetry. • Owned cross-functional execution across hardware, navigation, perception, and cloud components by driving milestones, dependency mapping, and risk/ escalation mechanisms, improving issue visibility by ~50% and reducing decision turnaround time by ~30%. • Drove key system-level decisions on platform selection and perception strategy, including shifting detection approach to improve success rate from ~20% to ~90%, stabilizing end-to-end system performance.

• Built a standardized post-evaluation workflow for a multimodal AI assistant, defining clear evaluation criteria across accuracy, safety, and compliance, and introducing structured error labeling to improve issue traceability. • Partnered with engineering to automate evaluation workflows using Azure Prompt Flow, Python, and SQL, improving evaluation efficiency by ~45% and reducing end-to-end workflow time by ~25%.