Project 20 · Generative AI
GenAI
PharmaAgents
Multi-Agent AI Research System
01 Abstract
Multi-Agent AI Research System
A multi-agent system on the Claude API where specialized agents collaborate on drug-research queries through task decomposition, tool use, and results synthesis.
02 Highlights
Methods & contributions
- Specialized agents — Literature Review, Drug Safety, Medicinal Chemistry, and Regulatory Intelligence agents coordinate on a shared task.
- Tool use — agents call structured tools and a coordinator decomposes queries and synthesizes results.
- Cheminformatics grounding — Lipinski, Veber, and Ghose drug-likeness rules implemented to literature values.
- Service layer — FastAPI with WebSocket streaming of agent progress.
03 Architecture
Data flow
query → coordinator (decompose)
→ [Literature · Safety · Chemistry · Regulatory] agents (tool use)
→ synthesis → streamed answer
04 Stack