jang1563/agentic-drug-discovery-system
Updated • 456
Auditable benchmarks for scientific agents: drug development decisions, calibrated abstention, tool use, and specialist-model reliability.
Note Public v0.2 includes CTDBench for benchmark execution and scoring, as well as an audited retrospective sickle cell disease vertical slice. The eight-stage platform remains a research roadmap.
Note Calibrated decisions based on labels derived from public sources, with abstention, temporal splitting, and explicit limits on autonomous use.