AI “Council of Models” Improves Workflows and Outcomes

To maximize the benefits of AI, biopharmaceutical manufacturers need to take an end-to-end systems engineering approach to their data, preparing it for AI while orchestrating the right models for each stage of a workflow. Relying solely on a single foundation model is often insufficient, particularly as complexity increases and therapeutics advance from pilot stages into production.

No single model excels at every task. Running the same engineering process through different AI models often produces significantly different results. Even repeated runs on the same model can yield inconsistent outputs, Farshid Sabet, CEO and co-founder of Corvic AI, tells GEN.

Variability may be acceptable for low-risk activities, but it becomes problematic when engineering diagrams, flow directions, operational relationships, and other complex data are involved. “Small inaccuracies can compound quickly,” he cautions, leading to unreliable results in production environments.

Corvic AI recently benchmarked leading frontier AI models against Corvic V5’s workflow orchestration platform, assessing their ability to extract piping and instrumentation diagrams (P&IDs) into XML files.

“For general text generation, today’s frontier AI models perform remarkably well and the differences between them are relatively small,” Sabet says. “But engineering workflows introduce an entirely different level of complexity.”

The benchmark found that relying on foundation models alone often resulted in inconsistencies, hallucinations, and poor repeatability. Corvic addresses these challenges by combining semantic data preparation with workflow orchestration that coordinates multiple AI models, validation steps, retrieval, and enterprise context to improve reliability.

Rather than replacing frontier models, Corvic’s platform integrates and orchestrates them, selecting the best model for each stage of a workflow based on the task, performance requirements, and cost.

Like wild horses

As Sabet says, “AI models are like wild horses. They’re incredibly powerful, but they need guidance, structure, and context before they can consistently solve complex enterprise problems.”

Typically, AI developers focus on improving the models themselves through training, fine-tuning, or prompt engineering while assuming enterprise data is already AI-ready. Corvic, instead, focuses on organizing enterprise knowledge through a semantic layer that enables AI systems to understand relationships across engineering documents, databases, diagrams, and operational systems.

“We work with the data independently of whether it’s manufacturing, chemistry, or biology,” Sabet says. “The data has to be organized in a way that allows AI models to recognize context and relationships. That’s the semantic layer.”

Once enterprise knowledge is structured appropriately, organizations can intelligently orchestrate multiple AI models throughout a workflow rather than relying on a single model for every task. Sabet refers to this approach as a “council of models,” where each model contributes its strengths to improve overall accuracy, repeatability, and efficiency.

A former Intel executive, Sabet founded Corvic AI to help organizations operationalize AI across complex enterprise environments. Today, the company works with manufacturers, life sciences organizations, and other enterprises to transform fragmented operational knowledge into reliable AI workflows that improve productivity and decision-making.

For organizations evaluating AI platforms, Sabet recommends looking beyond benchmark scores and considering three factors: how enterprise data is secured and governed, whether the platform intelligently matches AI models to different stages of a workflow, and how success will be measured through meaningful productivity outcomes.

To maximize the benefits of AI, Sabet reiterates, biopharmaceutical manufacturers “need to look at their data from a systems engineering perspective.

“The future of enterprise AI isn’t about finding one perfect model,” he says. “It’s about intelligently orchestrating enterprise data, semantic understanding, and specialized AI models into repeatable workflows that organizations can trust.”

The post AI “Council of Models” Improves Workflows and Outcomes appeared first on GEN - Genetic Engineering and Biotechnology News.