A collaborative AI partner for accelerating scientific discovery

Co-Scientist is a multi-agent AI system built to accelerate scientific discovery. Finding a genuinely novel research direction is one of the biggest bottlenecks to scientific progress. The system helps scientists break through it, generating, debating, and evolving candidate hypotheses to surface the few worth pursuing.

The design challenge was to make that multi-agent system feel like a genuine research collaborator. Beneath the surface, the system weighs thousands of directions against a grounded base of scientific literature, flags critical flaws, and reasons toward the strongest ideas. The product had to render that depth as something a scientist could direct, question, and trust.

As the Lead Interaction Designer on the team, I was responsible for taking Co-Scientist from zero to one. I helped shape the product direction and designed its core flows — a chat-based stage for refining a research goal into a full plan, human-in-the-loop controls to steer a run in progress, and a review experience for interrogating and iterating on the hypotheses returned. Validated with researchers across 100+ institutions and published in Nature, Co-Scientist launched as Hypothesis Generation within Gemini for Science, unveiled at Google I/O 2026.