Medical multi-agent system for clinical trial insights
Trusted answers, analysis, and exports—grounded in regulated data
For one of Europe’s largest pharmaceutical companies, Modulai built an anonymous, production-oriented multi-agent solution that answers questions about clinical trial outcomes by retrieving evidence, running analysis, and returning results in the right format—text, visualizations, or aggregated downloads.
Background
Clinical design data is spread across tabular trial databases, graph relationships, and unstructured reports. Teams needed faster access to design elements and supporting rationale, without manual cross-checking across systems. Any solution had to be traceable, compliant, and robust in a regulated setting.
Solution
We implemented a multi-agent workflow that plans the task, retrieves relevant sources, generates validated queries, and performs analysis before responding. The system returns open-format answers with citations to retrieved material, produces tables when appropriate, and enables data exports for downstream work. Guardrails enforce scope, reduce prompt-injection risk, and ensure results are compliant and aligned with regulatory standards.
Tools/Tech
Data was accessed through a combination of tabular stores, graph databases, and unstructured text retrieval. The graph layer enabled protocol relationship queries and comparison across related studies. Quality was maintained with automated evaluation in CI/CD and system monitoring, covering retrieval relevance, faithfulness, and end-to-end correctness—plus latency and cost.
