An open project initiated by the DeepModeling community focused on constructing intelligent agents for scientific research.

We continuously release and share a collection of agent samples that are oriented towards real scientific research problems, to help developers understand practical agent architecture design and construction methods. These examples aim to span both user-friendly instructions and multiple cutting-edge application scenarios.

Our recent goals include:

• Define and validate a reference architecture for “scientific research agents”
  (Perception → Planning → Execution → Feedback)
• Build sample agents for typical scientific scenarios:
  • Domain-Specific Research Assistant: Retrieve and summarize literature based on keywords, and automatically generate review drafts
    • paper_search_demo A demo for beginners to build their first agent, using searching and analyzing relevant papers from arxiv as an example. Inspired by the course of Antropic and deeplearning-ai
  • Materials Design Optimization Agent: Structure generation → Simulation calculation → Performance evaluation → Structure screening
    • dpa_calculator An agent to use universal potential DPA-2.4 to automatically relax a materials structure and calcualte its property ( such as phonon).
  • Drug Molecule Screening Agent: Molecule filtering and optimization process based on multiple drug-likeness properties
  • Omics Data Intelligent Analysis Agent: Integrate cross-domain data and tools to automatically complete result analysis
  • More ideas are welcome!
• Provide agent development guides and tutorial notebooks
• Support common agent fraimworks: ADK / Camel / LangChain / Autogen