A comprehensive research orchestration agent built with Google ADK. It uses a root_agent to coordinate multiple subagents, allowing you to search arXiv, parse PDFs, fetch citation graphs, search Semantic Scholar or Google, and look up conference submission rules all in one place.
To run the agent interactively in your terminal, use the following command:
adk run subagentsTo launch the agent's web interface, use:
adk web subagents-
Install dependencies: Ensure you have Python installed, then install the required packages:
pip install -r requirements.txt
-
Set up API keys: Depending on the tools configured (like Google Search or Semantic Scholar), you may need to set environment variables. Set them in your shell or a
.envfile before running the ADK.
arxiv_search_agent: Finds relevant papers on ArXiv.pdf_parser_agent: Extracts and summarizes content from PDF URLs or local files.citation_graph_agent: Maps backward and forward citations for papers.semantic_scholar_search: Explores academic literature.google_search_agent: Fetches recent news and discussions.conference_rules_agent: Retrieves formatting constraints for major ML conferences.
Once the agent starts, you will get an interactive prompt. You can provide instructions to instruct the root_agent to use multiple tools.
You can use the following comprehensive prompt to test the integration of all sub-agents:
"Search arXiv for the latest papers on 'Large Language Models Agentic Workflows'. Pick the most relevant paper and download/parse its PDF to summarize its key contributions. Then, fetch its citation graph to find backward and forward citations. If possible, use Semantic Scholar to gather more context on its authors, and run a Google Search to see if there's any recent news or discussion about it. Finally, check formatting rules for the next NeurIPS conference to see if this paper aligns with their page limits."
You can access the generated web interface (usually at a local URL like http://localhost:8000).
The UI allows you to chat with the agent similarly, but with a richer interface.