DrugPilot: LLM-based Parameterized Reasoning Agent for Drug Discovery

# DrugPilot: Performance Benchmark Results ## How to Run? ## Please visit our official platform: [https://www.drugpilot.xyz](https://www.drugpilot.xyz) ## Attention! Run with streamlit for a case !!!!! ## Key Features Parameterized Memory Pool: Standardizes diverse data types for efficient processing Feedback-Focus Mechanism: Real-time error monitoring and correction AI Model Integration: Seamlessly connects state-of-the-art drug discovery models ## Experimental Results Summary DrugPilot significantly outperforms state-of-the-art agents across multiple model architectures and task categories: ### Key Findings: - **Simple Tasks**: Achieves up to **99.8%** function accuracy and **98.0%** parameter accuracy - **Multiple Functions**: Maintains **98.7%** function accuracy and **96.3%** parameter accuracy - **Multi-turn Tasks**: Delivers **84.9%** function accuracy and **64.0%** parameter accuracy ### Comparison with Baseline Methods: DrugPilot consistently outperforms: - **ReAct**: Traditional reasoning-act framework - **LoT**: Least-to-most prompting technique - **CoT**: Chain-of-thought reasoning ## Performance Table - **Bold**: Best performance - Values show mean ± standard deviation --- **Note:** These results demonstrate static benchmark performance. For real-time interactive testing and full functionality, please visit our official platform: [https://www.drugpilot.xyz](https://www.drugpilot.xyz)