Key Innovation:
The Prompt Manager
Roko Labs and DHC reimagined their approach to AI driven regulatory writing with a new product called CentaurAI. Instead of scaling AI usage section by section, or relying on engineers to continually refine prompts, the teams built a flexible AI framework and internal tooling layer that placed control directly in the hands of regulatory subject matter experts.
At the core of CentaurAI is the Prompt Manager, an interface that allows DHC consultants to:
• View and edit prompts aligned to specific IND sections
• Define structured inputs and expected regulatory outputs
• Iteratively refine AI behavior based on domain expertise
Rather than acting as intermediaries, Roko Labs focused on enablement: building the systems, structure, and AI infrastructure that allow regulatory experts to drive outcomes directly.
A New Model
for AI Delivery
This approach fundamentally changed collaboration between the technical and the regulatory teams.
Instead of:
• SMEs explaining requirements to engineers
• Engineers translating those requirements into prompts
• SMEs validating AI output
DHC consultants could:
• Directly shape AI-generated regulatory content
• Experiment and iterate on IND sections independently
• Engage Roko Labs only for deeper technical challenges
“We realized pretty quickly that if we kept trying to do this section by section, we’d be working on this forever. The breakthrough was flipping the model: giving the client the keys. Instead of us being the bottleneck, we built the tooling and framework so their experts could actually drive the prompt engineering themselves. That’s what made it scalable.”
— Erin Ryan, SVP Product Management, Roko Labs
The result was a clear shift from dependency to expert empowerment.
The CentaurAI Platform
CentaurAI combines structured AI workflows with flexible expert control to support regulatory document automation at scale:
Modular Prompt Framework
· Each IND section is supported by structured prompts that combine general regulatory guidance with section specific logic.
Human in the Loop Review
· AI generates draft content that reaches approximately 85% completeness, while subject matter experts refine and finalize the remaining 15%.
Large Document Processing
· The platform supports summarization and synthesis of complex regulatory source materials, including documents thousands of pages long.
Cross Section Intelligence
· Dependencies between IND sections are incorporated into the workflow, improving consistency, traceability, and completeness across submissions.
Results & Impact
Efficiency Gains
• AI generated IND drafts reach ~85% completeness
• Significant reduction in manual regulatory drafting effort
• Faster iteration cycles across all IND sections
Operational Impact
• Increased regulatory throughput without increasing team size
• Reduced reliance on engineering support
• More time allocated to high value expert review and strategy
Strategic Advantage
• Lower operational costs
• Faster IND submission timelines
• Stronger competitive positioning in pharmaceutical regulatory consulting
Qualitative Outcomes
• Empowered regulatory subject matter experts
• Improved collaboration between domain and technical teams
• A scalable, sustainable model for AI adoption in regulatory writing
