This paper describes and benchmarks the competitor discovery component used within an agent-based AI system for rapid pharmaceutical asset due diligence. Given a specific indication, the competitor discovery AI agent searches for all drugs that comprise the competitive landscape for that indication and extracts standardized properties of these drugs. Competitor definitions vary across investors, data is paid/licensed, and fragmented across registries. Ontologies are inconsistent across indications, drug names have numerous aliases, are multimodal, and are rapidly evolving. While currently considered the best tool for this problem, LLM-based AI systems cannot reliably retrieve all competitor drug names, and there are no accepted public benchmarks for this task. To address this lack of evaluation, this paper uses an LLM-based agent to transform five years of multimodal unstructured due diligence notes from a private biotech VC fund into a structured evaluation corpus that maps indications to competitor drugs with standardized properties. Furthermore, we introduce a competitor validation LLM-as-a-judge agent that filters out false positives from the predicted competitor list to maximize precision and prevent hallucinations. In this benchmark, the Competitor Discovery agent achieved 83% recall, outperforming OpenAI Deep Research (65%) and Perplexity Labs (60%). This system is targeted at enterprise users; in a case study targeting a biotech VC fund, analyst processing time for competitive analysis was reduced from 2.5 days to approximately 3 hours (approximately 20x).
