Cancer Copilot
Cancer Copilot is an AI-powered tool developed by Color Health in partnership with OpenAI. It assists healthcare providers in making evidence-based decisions about cancer screening and treatment.
Closing gaps in cancer screening and diagnosis.
Cancer care is complex, time-sensitive, and highly individualized, and many patients face delays in screening and diagnosis due to evolving guidelines, fragmented processes, and administrative bottlenecks. Primary care providers often struggle to navigate personalized screening recommendations, while oncology teams spend weeks compiling the necessary diagnostic workups. Every delay impacts outcomes. Cancer Copilot was built to streamline this process, using AI to help clinicians interpret guidelines, identify care gaps, and ensure that patients receive timely, evidence-based treatment.
Applying AI thoughtfully in clinical care.
As Director of Product at Color, I drove the adoption of AI across the company and played a key role in strategizing, prototyping, and operationalizing Cancer Copilot along with a core cross-functional team. Our guiding principle was to be problem-driven, ensuring AI was applied not for the sake of AI, but as a tool to augment and support clinicians, streamline workflows, and ultimately improve patient outcomes.
AI strategy and prototyping.
I played a key role in defining and prototyping how AI could transform complex cancer screening and treatment guidelines into machine-readable formats. This involved working closely with OpenAI and our product teams to design AI-driven workflows that could translate dense, multi-page clinical protocols—often filled with intricate decision trees and care pathways—into structured data that could power AI-driven recommendations. To enhance the accuracy and usability of AI outputs, I worked on prompt engineering and content structuring, ensuring that clinical guidelines were interpreted correctly and aligned with best practices.
Model evaluation.
To ensure Cancer Copilot delivered clinically accurate recommendations, I worked with the clinical team and product team on the development of a golden dataset of synthetic patient cases, a critical benchmarking tool for testing and refining AI performance. I also collaborated with team to design evaluation criteria and define our strategy and processes for model evaluation. This approach allowed us to do continuous iterations on AI outputs and validate performance.
Responsible AI deployment.
I led Color’s transition from AI as an experimental tool to AI in production, ensuring that solutions were safe, reliable, and aligned with regulatory and ethical standards. I developed Color’s internal AI framework and governance processes, defining clear guidelines for responsible development and deployment. This work helped establish best practices across technical teams and laid the foundation for future AI-driven healthcare applications at Color.
Driving real-world impact.
Cancer Copilot has demonstrated measurable improvements in clinical workflows and is one of the earliest success stories about OpenAI models' applications in clinical care.
Copilot drives 4x increase in identifying missing diagnostic information.
Reduced analysis time from weeks to minutes, with clinical able to review patient records and identify gaps in just 5 minutes
Named one of Fast Company’s ‘8 Next Big Technologies in Health.’ and featured in WSJ.
One of OpenAI's first case studies and success stories in healthcare.
Images and video from OpenAI.com