LangGraph Research Assistant 🚀

Overview

This project implements a multi-agent research assistant using LangGraph, designed to retrieve, refine, and generate research-based responses. The pipeline intelligently selects sources (PubMed, Arxiv, Global), rewrites queries if needed, and generates a structured response.

---

Workflow

1️⃣ Query Classification

• The manager node determines the most suitable retriever based on the query type:
  • PubMed → For medical and healthcare-related queries
  • Arxiv → For AI, technology, and computer science-related queries
  • Global Retrieve → For general topics

2️⃣ Retrieval & Validation

• The assigned retriever fetches documents.
• If the retrieved documents are relevant, they are passed to the generate node.
• If not, the query is rewritten using the rewrite_query node and retrieval is attempted again.

3️⃣ Response Generation

• Once relevant documents are found, they are passed as context to the generate node.
• The final response is generated and returned.

---

Pipeline Diagram

The pipeline consists of the following components:

• manager → Classifies the query into PubMed, Arxiv, or Global Retrieve
• generate_pubmed / generate_arxiv / global_retrieve → Retrieves relevant documents
• rewrite_query → Rewrites queries if no relevant documents are found
• generate → Produces the final response

---

Installation & Setup

Prerequisites

• Python >=3.10
• Install dependencies:

pip install poetry
poetry install

export TOGETHER_API_KEY="your-key-here"