Scientific Researcher Agentic System

This repository implements a multi-agent system for an intelligent researcher tool. The system leverages a local Large Language Model (LLM) via Ollama to find and summarize academic papers in a sophisticated, multi-step workflow.

🏛️ Architecture

The system is built on a modular, hierarchical agent architecture, which makes it robust and easy to extend.

1. Researcher Agent: This is the master agent and the main entry point of the application. It understands complex, multi-step user requests and intelligently delegates tasks to the appropriate specialized agent.

2. Finder Agent: A specialized agent responsible for finding academic papers. It uses two tools:

   • ArXiv: To search for papers on the arxiv.org repository.
   • DuckDuckGo Search: For general web searches to find papers on other sites or to gather broader context.

3. Summarizer Agent: A specialized agent designed to summarize a given paper. It uses a tool to:

   • Fetch Web Content: Reads the text content from a paper's URL.
   • Summarize: Generates a concise summary covering the paper's objectives, methodology, and key findings.

Workflow Example

When given a task like "Find papers on multi-agent systems and summarize the first one," the workflow is as follows:

1. The Researcher Agent receives the task.
2. It first calls the Finder Agent with the topic "multi-agent systems."
3. The Finder Agent returns a list of relevant papers.
4. The Researcher Agent extracts the title and URL of the first paper from the list.
5. It then calls the Summarizer Agent with this information.
6. The Summarizer Agent fetches the paper's content and returns a summary.
7. Finally, the Researcher Agent presents the summary as the final answer.

---