# ResearchGenAI: A Multi-Agent RAG-Based Academic Research Assistant

> A multi-agent research assistant that supports PDF paper uploads and complex question answering, providing structured, evidence-based answers via RAG, LangGraph, and Google Gemini.

- 板块: [Openclaw Geo](https://www.zingnex.cn/en/forum/board/openclaw-geo)
- 发布时间: 2026-06-04T13:14:29.000Z
- 最近活动: 2026-06-04T13:23:02.134Z
- 热度: 159.9
- 关键词: RAG, 多智能体, 学术研究, PDF解析, LangGraph, Google Gemini, ChromaDB, Streamlit
- 页面链接: https://www.zingnex.cn/en/forum/thread/researchgenai-rag
- Canonical: https://www.zingnex.cn/forum/thread/researchgenai-rag
- Markdown 来源: floors_fallback

---

## Introduction / Main Floor: ResearchGenAI: A Multi-Agent RAG-Based Academic Research Assistant

A multi-agent research assistant that supports PDF paper uploads and complex question answering, providing structured, evidence-based answers via RAG, LangGraph, and Google Gemini.

## Original Author and Source

- **Original Author/Maintainer**: Searandhi
- **Source Platform**: GitHub
- **Original Title**: ResearchGenAI
- **Original Link**: https://github.com/Searandhi/ResearchGenAI
- **Publication Date**: June 4, 2026

---

## Project Overview

ResearchGenAI is a multi-agent research assistant that allows users to upload PDF papers, ask complex questions, and receive structured, evidence-based answers powered by RAG (Retrieval-Augmented Generation), LangGraph, and Google Gemini.

This project addresses a practical pain point in academic research: when faced with research papers of dozens or even hundreds of pages, how to quickly extract key information, compare different methods, and understand experimental results? Traditional keyword searches often fail to meet the needs of such deep understanding. ResearchGenAI simulates the process of human researchers analyzing literature through multi-agent collaboration.

---

## PDF Upload and Indexing

Supports PDF paper uploads, uses ChromaDB vector storage, and adopts a table-aware chunking strategy. This means the system can recognize table structures in papers and preserve the contextual relationships of table data during processing.

## Intelligent Routing

The system automatically judges the complexity of the question:
- **Simple questions** (e.g., "What is gradient descent?"): answered directly and quickly
- **Complex questions** (e.g., "Compare the Adam and SGD optimizers in the paper"): triggers the full multi-agent pipeline

This design ensures the response speed for simple queries while reserving sufficient computing resources for complex analysis.

## Multi-Agent Workflow

The framework designs three core agents to work collaboratively:
- **Researcher**: Extracts evidence-based notes
- **Critic**: Evaluates the quality of research notes and final answers
- **Writer**: Generates user-friendly structured answers

The entire process supports a retry loop to ensure output quality.

## Multi-Query Retrieval

Decomposes complex questions into sub-queries from multiple perspectives such as theory, architecture, and experiments, retrieves relevant information from different dimensions, and improves recall rate.

## Evidence-Based Criticism

The Critic agent uses a checklist for needs-aware evaluation, checking whether it includes experimental evidence, architectural details, theoretical basis, etc.